I’ve always had a deep appreciation for architecture: the geometry, the structure, the ideas behind it, as well as the history and cultural context of buildings in all their forms. Some of my old friends may still remember that I used to be especially interested in photographing buildings on the street, whether in the narrow alleys of Tsim Sha Tsui in Hong Kong or in the bustling downtown areas of Boston.

About a year ago, while I was still in school, I took an introductory course in the architecture department (ARCH 171). The class was essentially an introduction to what "Design" means for students interested in architecture. 9 years before this class, I was working as a freelance designer–doing UI, interaction design, and branding for small companies–and I thought immersing myself in design from scratch like this, should be a fun experience. It turned out to be exactly that. I genuinely enjoyed spending time with the younger peers and seeing the creativity that emerged during our presentations.

About two weeks into the class, I remember one afternoon after studio time when two professors came over to our group’s table. It was our first assignment. We were each assigned an existing architectural project by a well-known architect. Based only on observation, we had to draw architectural plans and then build a model that represents the idea/implication of the project. After briefly walking the professors through my process, they became interested in my draft, where I used a grid system to estimate and construct the curve of the building’s façade, essentially integrating curvature through a structured grid. They asked what year I was in, what my major was, and then said something along the lines of, “You should study architecture.”

I was very, very inspired by that moment. At one point in my life, I truly did dream of studying architecture. I still remember traveling along the East Coast about ten years ago, photographing buildings designed by my favorite architects–works by Le Corbusier and I. M. Pei around the Cambridge area. Although my life path has taken a different direction, simply being able to appreciate how powerful and beautiful architecture can be still means a lot to me. And for now, that feels like enough.

Now let's get to the work. All of them shown here was created about a year ago, and the only reason I decided to write these thoughts down is because I genuinely loved what I designed, and I did it from scratch. There were no AI tools involved, just me, my thoughts, and my passion. I wanted to document it here as a small homage to a dream I once had.

The class itself offered a glimpse into what an architectural education feels like, especially at a school known for its strong architecture program. The two major assignments were structured around a midterm review and a final review, and each required presenting our work in front of both students and professors. I really enjoyed this in-person presentation style.

Midterm Review

Final Compendium

I was excited to share my work at the end of the semester. During a group presentation where all students’ projects were displayed in the lounge of Temple Buell Hall, I felt that being an architecture student is something quite special. The environment was formal yet enjoyable. Architecture sits at the intersection of art, design, and engineering, and that combination gives it a strong sense of professionalism and purpose.

Let me briefly walk you through my process for this miniature project. We were asked to design an observation tower for one of three sites in the University Arboretum near the Japan House—a peaceful and picturesque area. I chose Site C, located south of the Japan House, with lush greenery and an open corner where I placed my tower. I used to live near the arboretum and visited it often, so I felt a personal connection to the site.

Inspired by the traditional hexagonal gates of Suzhou Gardens, I wanted the design to blend naturally with its surroundings. Each floor of the tower rotates 90 degrees, offering distinct views:

• the west grove and Japan House parking lot,
• the south grove and Pollinatarium,
• the Orchard Downs housing community to the east,
• and the Japan House and Idea Garden to the north.

The open ground floor creates an inviting entry, while the rooftop offers panoramic views. Visitors ascend via stairs or an elevator, and the structure is supported by four main pillars.

The idea behind this project aligns closely with my love for wandering and exploring spaces. Walking through Area C gave me a strong sense of connection to the natural elements—trees, wind, and weather—which directly shaped the design approach behind my work. As "Walking as Research Practice" beautifully states: “We knead, imprint, trample, and leave traces, whether we want to or not.”

When I finished school last summer and moved back to the Bay Area, I brought this model with me on the airplane, carefully wrapped in foam. The good news was that it’s still intact, and now it sits on my desk every day.

Maybe someday, it will motivate me to design something bigger.

This board game is the result of my coursework in INFO 416 during Spring 2025 at the University of Illinois. As a climbing enthusiast, I was able to put my thoughts and the game design concepts I learned from the class into practice and create an actual physical game from scratch. I really appreciate that people enjoy the idea and the game itself!

Creating the Game

At the ideation stage of designing the physical game, I knew right away it had to be about climbing. Climbing has been such a big part of my life in the past year, and I felt like it was an area of culture that hadn’t really been explored much in board games. What wasn’t clear to me at the start, though, was how exactly that would look. I had once imagined building a digital climbing gym simulation game—something where players could either be climbers or gym owners, setting routes and managing the space. But as I thought more about what kind of game would actually be fun to play with friends around a table, it made more sense to center the experience around the climber: their progression, the gym environment, and the quirky moments that make climbing so community-oriented and personal.

From the start, I wanted the game to be both fun and grounded. I designed it as a kind of simulation of the bouldering gym experience. I filled the cards that the player drew with climbing jokes and references—like losing chalk to someone else in the gym, or gaining stamina by drinking Yerba Mate, which, if you’ve spent time in any climbing gym, you know is basically part of the culture.

While refining Boulder Quest, I focused on several key aspects to improve both gameplay and player experience. One of the main priorities was making sure the game remained accessible to newcomers while still feeling authentic to climbers, which meant carefully balancing the use of climbing terminology. I also spent time adjusting the overall game balance—especially the relationship between resource gain and loss—to keep the pacing engaging without making it overly punishing. I also considered the physical usability of the components; I paid close attention to how resources were tracked, how tokens were placed, and how intuitive the player boards felt during gameplay. These areas became central to my iteration process among all the concerns

Iterating the Game: Feedback & Challenges

Throughout the semester, I held four rounds of playtesting, and every session gave me something new to think about. One of the most consistent themes in the feedback was about resource balance. Players pointed out that the early versions of the game had too much loss and not enough gain—especially from the gym and training cards. Even though people found the flavor text funny, the high frequency of losses made the game feel punishing. It discouraged players from drawing those cards sometimes, which wasn’t what I wanted.

To fix this, I adjusted the math behind the cards. I increased the frequency and value of gains and softened the consequences of losses. Behind the scenes, I created a spreadsheet to simulate various draw combinations and balance outcomes across different situations. I want to make sure the game always felt like it was offering progress, even when things didn’t go perfectly.

Feedback also helped me realize that tracking resources needed to be more visible and intuitive. Originally, I didn’t even include a section on the climber’s board for skill tokens. I thought it might simplify the game to have fewer tokens to move around. That turned out to be a mistake. Without a clear place to track skills, players kept forgetting what they had, and the extra mental load made the game less enjoyable. In the final version, I added skill slots directly onto the board—and that change was well-received. It made everything more transparent and gave players a stronger sense of control over the resource allocation.

Another issue that came up during playtests was unclear placement of tokens. The layout of the climber’s board sometimes caused players to confuse where strength, skill, stamina, or chalk tokens were supposed to go. I’ve brainstormed a few potential solutions—like redesigning the board with clearer zones or using icon-based trays—but for now, I’ve included a cheat sheet and a “spectator” role to help smooth over the learning curve. The spectator doesn’t play, but they help distribute tokens, keep track of rules, and speed things up. That role ended up being helpful in longer sessions.

Aside from gameplay issues, one of the biggest challenges I ran into was producing the physical prototype. Digitally, the design process went smoothly. I built a Python script that allowed me to feed in card data from a JSON file and automatically generate all the visuals I needed. That saved me a lot of time and gave me the flexibility to make changes quickly when balancing the cards. But printing them was a different story. I couldn’t find a campus printer that could do double-sided prints on thick cardstock, so I had to get creative. I manually aligned front and back layouts, printed them out, and then glued them onto pre-cut cardboard pieces. It wasn’t the prettiest or easiest solution, but it gave the cards enough weight to actually feel good in players’ hands.

Another tricky part was managing the number of stages in the game. At first, I had three stages with different climbing grades, but that made the game drag on too long. In one session, we only played stage 1 and it cost about 30 minutes. So I restructured the stages into just two, with each wall stacking easier problems on top and harder ones underneath. It gave the game a natural sense of progression without overstaying its welcome, and the total playtime now sits at a comfortable 1 to 1.5 hours.

In the end, each challenge—whether it was a design decision or a production hurdle—became a part of the game’s story. Boulder Quest evolved not just because I had a vision, but because other people played it, questioned it, and helped me see it from different angles. That feedback loop, that collaboration, is a lot like climbing itself: you try something, you fall, you learn, and then you try again—this time a little smarter.

Vision & Shortcomings

From the beginning, my vision for Boulder Quest was to create a board game that captured the essence of indoor climbing—not just the physical challenge, but also the community vibe, the in-jokes, and the feeling of progression that climbers experience session after session. I think I was able to achieve that balance in many ways. The humor landed well during playtests, the theme resonated with climbers, and the pacing—after several adjustments—felt satisfying without being too heavy.

That said, there are still areas where the game fell short. Resource balance, while much improved, isn’t perfect. For example, in the final playtest, players struggled to collect specific resources like “body work” during the harder stages, which led to some stalling. It wasn’t a game-breaking issue, but it showed that certain card distributions still need more variety and flexibility. Physical accessibility is another area that needs more work. Token management, in particular, was more tedious than I’d like—it took up more time and attention than it should, and as the “spectator,” I found myself doing a lot of manual cleanup just to keep the game moving.

These are things I plan to address in future iterations, but I also see them as natural growing pains in a project that started from scratch and evolved through trial, error, and a lot of learning.

Core Values & Intended Experience

I wanted the game to reflect the idea that climbing isn’t just about reaching the top—it’s about the small wins, the fun moments, and the learning portions. That meant designing a game where players of different skill levels could enjoy themselves, where setbacks didn’t feel punishing, and where the gameplay encouraged both individual strategy and light social interaction. The character roles were built to reflect different climbing styles and personalities, without creating a strict power hierarchy.

The play experience I envisioned was something calm but engaging—strategic but never stressful. I wanted each turn to feel like a meaningful choice: whether to rest, push through a problem, or train for something harder. I also wanted to recreate the rhythm of a real climbing session, where difficulty builds gradually, and where players feel themselves getting better—or at least more prepared—as they go. I wasn’t aiming for intense competition; instead, I focused on friendly tension, where players can still cheer each other on, laugh at the card effects, and share a story by the end of the game. If the table ends up joking about someone hoarding chalk or trying to send a V6 with no stamina left, then I’ve done my job right.

Looking Forward: Improvements, Accessibility & Audience

There’s still plenty of room for Boulder Quest to grow in future iterations. One of the first things I want to improve is the game’s flow and progression mechanics. Right now, there’s a loophole where players could avoid climbing entirely by just drawing gym and training cards, which slows the game down and breaks the intended rhythm. I’d like to introduce incentives for attempting problems regularly. I also plan to diversify the card requirements—especially in the higher grades—so that players don’t get stuck searching for one specific type of resource. Token placement and resource tracking are still more tedious than they should be, so I’m exploring better board layouts and maybe even component redesigns to make the experience smoother. Down the line, I’d love to add random events, asymmetric character abilities, and more social mechanics to better reflect the playful unpredictability and collaboration in real-world climbing.

In terms of accessibility and approachability, I’d say the game is moderately accessible and approachable. It uses familiar mechanics like turn-taking, resource management, and character-based strategy, which helps new players get into it quickly. The theme is clear, the player aids are useful, and most people pick up the flow of the game within the first few rounds. That said, the game does have a lot of physical components—tokens, icons, boards—and that can feel overwhelming at first. I’ve tried to make that more manageable with things like the cheat sheet and the optional spectator role, but I know there’s more I can do to make the game less fiddly without losing its depth.

As for the intended audience, Boulder Quest is designed for teens and adults aged 12 and up, especially those with an interest in climbing or casual strategy games. I chose that age range partly because of the laser-cut wooden components, but also because the game requires a bit of planning and attention to detail that younger players might not enjoy as much. It’s meant for small groups—2 to 4 players—like friends at a climbing gym or people looking for a chill, thematic game night. I’ve tried to make it welcoming for both climbers and non-climbers by blending familiar cultural references with intuitive design. Ideally, the game feels like a relaxed climb session: a little competitive, a little chaotic, and something everyone walks away from with a good story.

Playtest Details & Thanks!

The current Boulder Quest has gone through 3 version of iterations:

• Version 1.0: first prototype
• Version 2.0: updated prototyped based on 1.0
  • based on the feedback from the first few playtests
• Version 3.0: final prototype

Here are some of the details for each playtest sessions:

• May 7
  • Playtesters: Dave, Zhiheng, Lizi, and Shilan
  • Play time: roughly 1.5 hrs. finished the whole game and have winner.
  • Where: Urbana Boulders
  • Version played: version 3.0
• Feb 26
  • Playtesters: Min, Aaron, Josh, and Daniel
  • Play time: roughly 40 minutes. Didn’t start stage 2.
  • Where: Urbana Boulders
  • Version played: version 2.0
• Feb 22
  • Playtesters: Yiyu, Joe, and Tommy (myself)
  • Play time: roughly 40 minutes.
  • Where: Urbana Boulders
  • Version played: version 1.0
• Feb 20
  • Playtesters: Julia, Jackson, Danny, and Yuge
  • Play time: roughly 20 minutes. Didn’t finish stage 1
  • Where: Armory 182 in class
  • Version played: version 1.0

👀 There will be more!

It has been almost five months since I started bouldering. I thought it might be a good chance to recollect my past recorded videos and see how I did throughout this wonderful journey. I want to share my top 10 favorite climbs since I started. There are definitely great routes I didn't have the chance to record, and I enjoyed those moments mingling with people even when the camera was not on.

No. 1: V0-2

Recorded on Feb 27, 2024

This is the very first recorded send I did back in February, when I was just about three weeks into the bouldering world. I actually tried this route the day before but was not able to finish it. I'm not sure if I was exhausted or if I just didn't figure out the right postures. This send was recorded in the morning, and I hadn't eaten anything besides some breadcrumbs from a sandwich I bought at a coffee shop. I had no preparations whatsoever, and when I chatted with my friend Litong about this route, I just went for it. To my surprise, I finished it on the first attempt (for that day). Then I asked her to help me record this video for the second attempt—it obviously worked again. This route is not exactly energy-consuming despite it is a quite circuitous path. On the fourth-to-last boulder, I realized the importance and effectiveness of flagging on the wall to save energy since it kept my center of mass closer to the wall. Techniques matter! Anyway, this is definitely worth remembering. Thanks, Litong, for the video, and again, for taking me to this climbing journey!

No. 2: V2-4

Recorded on Apr 16, 2024

I finished this V2-4 with pride. No joke, the first few times I tried it, it was too hard for me to stay on the route at the horizontal pinch (the one just above the light green boulder) because I didn't match on it. I thought I had to train harder on my biceps to be able to do this. Then I left this route and went for another V0-2 overhang (the yellow boulder just to the right of this blue route), and only campused on that for almost two weeks. Although it helped, I still think it was mainly because I didn't match on the pinch. Well, I finished this in a session with my friend Nick, where I chatted with him about my uncertainty in finishing it. Oh my, the look on my face was pure joy when I got off the wall. I persistently tried this route again and again at the start of each session since it made me feel like I had the ability to climb outdoors. Meh, no, I was dillusional for sure—but that doesn't I'm not able to do it indefinitely. Allez! Thanks, Nick!

No. 3: Sandbagged V6-8

Recorded on Apr 21, 2024

It's really satisfying to have a sandbagged route that I can get my hands on in my local gym—and why I have to rank my local gym as the best gym in the world. I said satisfying, not just because I was able to try it, but also because, as Jed, the route setter, said, it gave me the incentive to try harder boulders. Once this set was finished, I gave it a try and realized, "Hey! I might be able to finish this V6-8." I never realized that my core strength was good enough to finish it, but yay, I do have good control over my core along with my lower body flexibility.

No. 4: V4-6

Recorded on Apr 25, 2024

This was a pretty random route Nick and I chose to project on a sunny afternoon. We originally wanted to work on another cave route, which is the same grade (V4-6) as this one. Eventually, we gave up after more than 15 attempts because we couldn't figure out the good start for that one. This was probably the first time I felt great about using a volume as a hold to pull myself up to the crimp right after the corner (above the edge).

No. 5: V2-4

Recorded on Apr 27, 2024

It's a funny V2-4 that exploited my middle fingers for days. The key was trusting my knee to generate the momentum needed to reach an upper position, allowing me to place my left foot in the next pocket, which is about one meter away. Also, the end of it is trash. Tall people can just reach it, but as a short guy, I have to trust my jump and the precision of my hand placement on the final pocket at the top. The movement looks cool though! Thanks, Dave, for helping me record the video!

No. 6: V5

Recorded on May 15, 2024

After my spring semester ended, I headed over to California to spend my break. Benchmark Berkeley was the place my friend Falcon introduced me to, and I had a lot of fun climbing there. It definitely has much more facility than Urbana Boulders, my local gym and also the gym where I started climbing. It has a MoonBoard, bunch of gym facilities, some yoga gadgets, several hangboards, and a diverse range of routes with fixed grades (instead of a vague range) marked at the start of each problems. I know some people may say grades are just numbers–I do value them since I'm just a beginner/intermediate(?) climber, and grades definitely help me see whether I've progressed or not. But hey, if I want to climb a V10 in a random tomorrow, I would say it's for fun—because why not? I climb, therefore I have fun.

Random thoughts aside, this V5 was one of the few routes I figured out myself without any betas from others. Additionally, it has the dynamic vibe I like and helped me practice my slab skills. The heel hook on the second-to-last sloper was a spontaneous move. Skills, skills, skills! Thanks, Falcon, for his guest pass and all the fun climbing time we spent there.

No. 7: V4-5

Recorded on May 20, 2024

Mosaic Boulders is a relatively small bouldering gym very close to the UC Berkeley campus—just two blocks away from the famous Sather Gate. I'd say it's probably the same size as Urbana Boulders. But that's what I love most about small gyms near schools: I get to meet new friends without any pressure. We share similar vibes and are willing to open up to exchange our thoughts without mental barriers, at least that's how I view it.

This is the first-ever "dynamic" route I finished. While it's graded a V4-5, only the start is the hardest. The rest is pretty straightforward. Before I attempted it, I saw a girl try it about 50+ times. I think that was one of the reasons that motivated me to give it a try. It's about perseverance and relentless attempting!

No. 8: V4-5

Recorded on Jun 25, 2024

I spent 10+ attempts on this route before finishing it. Initially, I was trying, like others, to catch the medium-sized jug on the volume, which requires a lot of momentum to boost myself up there. Then I thought, no, there has to be an easier way. So I considered more static moves, and the result was using the one boulder in between (the small sloper) to support me and get closer to the medium-sized jug, which was much easier. I like how I did it in a relatively static way.

No. 9: V4-5

Recorded on Jun 25, 2024

This route is a bit random and came about the other day when I was looking for new routes to work on. The start is definitely a bit dynamic, but I think the key is precision on the crimps on both sides of the volume. After less than 10 attempts, I was able to finish it with confidence in jumping at the start. It looks cool. Thanks, Nate, for recording the video!

No. 10: V6-7

Recorded on Jun 30, 2024

Oh my, oh my, this is my first (probably actually legitimate?) V6-7. It's this boulder that made me believe that I have indeed been making progress in climbing. The first move is a toehook on the big jug while catching the upper side of the next big jug. Then the next move is to jump to the crimp, landing with my foot. What's really fascinating is how my climbing friends, who gave me the beta, suggested using a "chest" or "shoulder" bar to lay back on the second-to-last boulder in order to reach the top of this route. It's not possible to grab the top boulder without this move, since the top part is smooth and offers no friction. A good problem, and I guess it marks nice progress for me as well!

I want to conclude this post with some notes, things I did and routines I had in mind throughout my climbing sessions:

• Climb the same finished route a bit more to discover new possibilities, train my intuition to apply techniques spontaneously, and gain confidence for tackling harder routes.
• Do longer warm-ups before my climbing sessions: 10+ minutes consisting of (power or not) pull-ups, stretching the entire body (feet, neck, hips, wrists, ankles, etc.), hangboarding on jugs for a sufficient amount of time. And of course, climb on easy routes (V0-2) and harder routes that I've finished before tackling real problems.
• Have more regular training session! Having solid muscles, great flexibility and endurance is always better.
• Talk to strangers and share thoughts! One special thing I've gained from climbing is building trust, and I mean that in two ways. The first kind of trust is about trusting myself and every movements, especially the seemingly risky ones. The second kind of trust is about being kind and nice to the people I interacted with. This has helped me a lot in terms of having in-person social interactions, especially since I have an introverted personality. People are beautiful. Trust climbers around you.
• Take proper rests! Honestly, I didn't take enough rests between sessions because, like most climbers, stopping climbing is the hardest part. But hey, recovery is important. Resting during sessions also helps me take some time off, talk to people instead of feeling the pressure of not being able to solve a project, and relax my body and muscles. I just love climbing
• I still have not yet buy a brush. And until now, it becomes necessary for me to use brush on some harder routes. Brushing for others is also a good manner 😉
• Record my sends with proper notes. This helps me remember the techniques I applied on each send and may consolidate my memory to prepare myself for future project tackling. I designed the following climbing journal page, printed several pages, and clipped them onto a lap board of A6 size. It's portable, and I can use a pencil to freely draw anything I want. The only recent drawback I've found is that I always feel a bit shaky after my send due to my torn muscle.

Keep climbing and keep having fun!

As per my last post, I've been (indoor) bouldering for about 2 months now. I can now say I'm pretty confident in topping the red-taped routes (V0-2), and able to finish almost half of the yellow-taped routes (V2-4). To my surprise, I was even able to finish a V4-6 the other day; well, not sure if counts since I asked for a Kevin for the beta.

I started bouldering on Feb 6 this year and noticed that there's a real-time person counter widget displayed on my gym's website. With this convenience, I realized that I could write a Python script to extract the webpage parse the person count, and record it in a fixed duration for each opening day. So I went ahead and wrote one, and let it run in the background via PM2 on my laptop (as I never turn it off, it naturally becomes the first choice, instead of running on a server). A few weeks later, after spring break, I've got about 1 month of data; now almost 2 months of data points. Visualizing them corresponding to each date on 7 different subplots on Jupyter, it looks like this:

As you can directly observe from these subplots, on weekdays, except for Wednesday, the trend for the opening hours is pretty much the same. It makes sense because during the low hours, most students and faculty who visit here are either in classrooms or offices. Wednesday nights are usually members' night, so you see that during those hours, the data points hit very high; while there are many people, I went there one day and it is surprisingly "not very crowded." In terms of the availability of the routes, people are just very nice, and, well, you have to take some rests between each climb.

I used to go to the gym based on this chart and scheduled to go there when there weren't many people ($\leq 20$). It does feel really great when you're one of the only few climbers in a certain area with walls of routes that you are free to climb and record a video about. However, after that Wednesday members' night I went to, I realized having people around isn't actually a bad thing; at least you have someone to talk to, potentially getting beta for some routes I've tried many times. Recording videos can be handy even if there are quite a few people as well; most climbers here are so nice, and whenever I asked them, they just helped me. So right, this visualization is now purely for fun.

I also had other plots; this one is a bit fancier, as it has some background colors indicating each time window, which provides some kind of intuitive insights on how many people are there.

I thought about using Markov chain or time-series models to potentially predict the possibility of a time range being rush hour or not. While this goal is somewhat undeterministic, as a person who has visited there quite often (now more than four times a week), I still want to give it a try and at least achieve some triumphs in modeling this problem. I had some observations when thinking about the possible relevant events that could cause the number of people to change significantly:

• Change of routes: Whenever the gym route setters make big changes to the routes, I notice a relative surge of some groups of intermediate/advanced climbers hitting the gym, discussing, and having fun.
• Some other events that could possibly have an impact include:
  • Bad weather, low temperatures: fewer people visiting
  • Special events on campus: fewer people visiting
  • Exam weeks: fewer people visiting
  • Spring/fall/winter/summer breaks, holidays: fewer or more people
  • Members' night: more people visiting

And that's it; these are my current thoughts about this little toy "project" I've had running in the background for a while. I'm still thinking about the computer vision side of bouldering tutoring, using NeRF to reconstruct the 3D scene, and allowing analysis of the boulderers' movement, as well as possibly recording which routes they have finished in an interactive way. It may just be for fun and not practical, but why not? It's a cool idea.

At this point, I'm glad that I started the bouldering journey in this relatively small place but with good routes and nice people around. It's a warm place to visit.

Here, I present to you some of my favorite routes recently:

So when is the best time to visit the bouldering gym? Anytime works, as long as the place is welcoming. Allez!

I. Motivation

In project 3 (a project on the topic of gradient domain editing), we implemented texture style transfer, learning how to transfer the "drawing style" from one picture to another based on pixel intensity. Given the creative perspective of this notion, we want to incorporate additional techniques not yet covered in class projects, such as object averaging (morphing), image warping, matting, and hole-filling, to create a pipeline for “Face Migration in Classical Artwork.”

Our inspiration comes from a creative rendition of Vermeer’s “Girl with a Pearl Earring” (shown in Figure 1), where the girl's face is replaced with that of an oil-paint cat. This gave us the idea of seamlessly merging objects from one classical artwork into the scene or objects of another. Our primary objective is to facilitate face-to-face swaps that do more than just blend faces; we also want to apply texture transfer to match the drawing style and morph only into the face region, preserving the integrity of the original artwork. Figure 2 shows another example we aim to achieve, swapping the two faces in “American Gothic” by Grant Wood.

II. Approach

Our approach comprises several steps, each of which is covered in the class. The pipeline shown above provides an overview of our method (refer to Figure 3), and here are steps with more detailed descriptions:

1. Image Selection: We first obtain two sets of images. One set consists of pictures of a "cat," the other set includes numerous classical portrait artworks.
2. Source and Target Images: Select one image from each set, labeling them as the 'source' and 'target' images, respectively.

3. Facial Landmarks Extraction: Extract facial landmarks from both images using the standard 68 landmarks model.

• For human faces, we automatically extract the landmarks using dlib’s detector and predictor, which is a mature technology.
• For cat faces, we manually extract the landmarks using a Python tool we developed, taking the positions of dlib’s predictor landmarks as reference (Elmahmudi, Ali, et. al.). For more details, see the README file in the code repo.

4. Triangulation: Triangulate the facial landmarks of each face to match the same mesh structure. This is achieved using a convex hull and the Delaunay algorithm.

5. Affine Transformation: Perform an affine transformation from the stylized source face to the target face, or vice versa. Obtain the mask of each face.

6. Poisson Blending: Apply Poisson or Mixed blending using the source face's region of interest, the target artwork, and the source face mask.

III. Results

IV. Implementation

For this project, Python was our primary programming language, complemented by a range of open-source libraries. Below is a list of these libraries in our project:

• cv2 (OpenCV): 
  • General Functions: Used for reading images, converting color spaces, visualizing landmarks on images, and calculating the bounding box coordinates.
  • Convex Hull and Subdiv2D: Used for creating convex hulls and Delaunay triangulation, essential in the process of meshing facial landmarks.
  • cv2.getAffineTransform(), cv2.warpAffine(): These functions perform affine transformations for morphing faces from source to target images.
  • cv2.fillConvexPoly(): Fills a convex polygon, which we used for masking and blending regions in the image.
  • Poisson Blending: we opted for the cv2.seamlessClone() function instead of the functions built during project 3 for its efficiency and effectiveness in the blending step. 
• dlib: Used for its robust facial detector and predictor, providing the facial landmarks necessary for our transformations.
• Pillow: Used for opening and handling the content and style images from their respective directories.
• Others: matplotlib, numpy, json

For image datasets, some were acquired from the internet (e.g., Google Arts, Google Search), while others were generated using GAN technology (e.g., DALL-E). The sources of these images will be listed in the reference section.

V. Problem

The main problem was achieving a seamless blend of the source face with the target artwork's color palette and style. We initially tried mixed gradient blending, which sometimes caused the source face to lose its unique features. We then experimented with poisson gradient blending, which retained more of the source face structure and was ultimately chosen for our project. We provide users with both blending options, allowing them to select based on their preference.

For future improvements, we hope to explore other techniques that account for variations in lighting, saturation, and color profiles. Potential solutions include:

• Implementing a multi-scale blending approach for preserving finer details while ensuring a smooth transition. 
• Incorporating color correction algorithms to match the source face's color tone with the target artwork.
• Enhancing triangulation, especially around key facial features, to achieve more detailed and accurate morphing.

Our initial goal of facilitating general object migration between artworks, time constraints led us to focus solely on facial swaps. An automatic segmentation method could be incorporated into the project to handle the different shapes, colors, and textures present in various objects, which would be quite interesting to observe as objects migrate through each other across different artworks. Recent developments with Semantic Automatic Matting (SAM) (Kirillov et al., 2023) seem to be a promising option to be integrated in the future.

VI. Acknowledgements

Tommy Liu

• Project report writing and the finalized version
• Implemented a tool to manually annotate the 68 landmarks for cat face, as well as save/load the landmark file to be used in the Notebook
• (Main notebook) implemented the parts for landmark extraction, triangulation w/ convex hull and Delaunny algorithm, face blending, the final output and comparing section, as well as the final code integration
• Ran demo on various source and target images to showcase the project's capabilities
• Wrote the README file in the GitHub repo

The full code is located here on GitHub.

Reference

List of images used

• van Eyck, J. (c. 1438). Portrait of Giovanni di Nicolao Arnolfini. [Painting].Berlin.
• Christus, P. (c. 1465-1470). Portrait of a Young Girl. [Oil-on-oak panel]. Gemäldegalerie, Berlin.
• van den Valckert, W. (1617). Man With a Ring. Location Unknown.
• Vermeer, J. (c. 1665). Girl with a Pearl Earring. [Oil painting]. Mauritshuis museum, The Hague.
• Repin, I. (1884). Portrait of Vsevolod Mikhailovich Garshin. [Oil on canvas]. Metropolitan Museum of Art.
• Wood, G. (1930). American Gothic. [Painting]. Location Unknown.
• Elmahmudi, A. and Ugail, H., 2021. Identification of facial landmarks using Dlib [Digital Image]. In A. Elmahmudi and H. Ugail, "A framework for facial age progression and regression using exemplar face templates," The Visual Computer, vol. 37. doi: 10.1007/s00371-020-01960-z. Available: https://www.researchgate.net/figure/dentification-of-facial-landmarks-using-Dlib-a-Facial-landmarks-b-The-position-and_fig2_343699139
• Adam88xx. (2014, May 23). Cat Kitten Gaze Regard Glance Look Head. [Digital Image]. IStock. Available: www.istockphoto.com/photo/cat-gm493250135-40361554. Accessed 6 Dec. 2023.
• Aleksandrov, V. Red Cat. [Digital Art]. ArtPal. Available: www.artpal.com/VladimirAleksand?i=238586-12. Accessed 6 Dec. 2023.
• Brooks, K. (2015, Sept. 3). Cats Are Taking over Famous Western Artworks and We’re Definitely Not Mad about It. HuffPost. Available: www.huffpost.com/entry/cats-are-taking-over-famous-western-artworks_n_55e75737e4b0aec9f355c018. Accessed 6 Dec. 2023.
• Flood, S. Orlando. [Painting]. Year and Location Unknown.

List of readings

• A. Kirillov, E. Mintun, N. Ravi, H. Mao, C. Rolland, L. Gustafson, T. Xiao, S. Whitehead, A. C. Berg, W.-Y. Lo, P. Dollár, and R. Girshick, "Segment Anything," 2023. arXiv:2304.02643 [cs.CV]. [Online]. Available: https://arxiv.org/abs/2304.02643
• D. King, "Dlib's demo code for using 68 landmarks predictor," dlib.net. [Online]. Available: http://dlib.net/face_landmark_detection.py.html. [Accessed: Dec. 05, 2023].
• OpenCV, "seamlessClone() - OpenCV 3.4 documentation," docs.opencv.org. [Online]. Available: https://docs.opencv.org/3.4/df/da0/group__photo__clone.html#ga2bf426e4c93a6b1f21705513dfeca49d. [Accessed: Dec. 05, 2023].
• OpenCV, "Subdiv2D() - OpenCV 3.4 documentation," docs.opencv.org. [Online]. Available: https://docs.opencv.org/3.4/df/d5b/group__imgproc__subdiv2d.html. [Accessed: Dec. 05, 2023].
• L. R. Finka et al., "Geometric morphometrics for the study of facial expressions in non-human animals, using the domestic cat as an exemplar," Scientific Reports, vol. 9, no. 1, Art. no. 9883, Jul. 2019, doi: 10.1038/s41598-019-46330-5.
• G. Martvel, I. Shimshoni, and A. Zamansky, "Automated Detection of Cat Facial Landmarks," 2023. arXiv:2310.09793. [Online]. Available: https://arxiv.org/abs/2310.09793 [cs.CV]. [Accessed: Dec. 05, 2023].
• G. Martvel, N. Farhat, I. Shimshoni, and A. Zamansky, "CatFLW: Cat Facial Landmarks in the Wild Dataset," 2023. arXiv:2305.04232. [Online]. Available: https://arxiv.org/abs/2305.04232 [cs.CV]. [Accessed: Dec. 05, 2023].
• Tech4Animals Lab, "CatFLW: Cat Facial Landmarks in the Wild Dataset," Tech4Animals, 2023. [Online]. Available: https://www.tech4animals.org/catflw. [Accessed: Dec. 05, 2023].
• Piętka, Ola. “Face Swapping & Style Transfer.” GitHub, 14 Nov. 2023, github.com/OlaPietka/Portrait-Style-Transfer. Accessed 7 Dec. 2023.
• Q. Zhang, "Landmark Detection for Animal Face and 3D Reconstructions," zhangtemplar.github.io, 2023. [Online]. Available: https://zhangtemplar.github.io/animal-keypoints/. [Accessed: Dec. 05, 2023].
• S. Mallick, "Face Swap using OpenCV," learnopencv.com, 2023. [Online]. Available: https://learnopencv.com/face-swap-using-opencv-c-python/. [Accessed: Dec. 05, 2023].

On a cold night, as pedestrians hurriedly traversed the street, Gereon's attention was drawn to an advertisement on a pillar. The inscription on the poster read, "Jeder einmal in Berlin!" which in English means "Everyone once in Berlin!" In one of the Chinese subtitle translations, "每个人都应该来柏林," and in English again, is "Everyone should come to Berlin!"

This scene is from the German TV series "Babylon Berlin," where Gereon Rath is the male protagonist. I started watching it while taking a German popular culture class, which spurred my curiosity to learn more about Germany. Gereon, a WWI soldier suffering from wartime PTSD, later became a police officer in Berlin, either to heal or escape from his past. Throughout the first three seasons, viewers witness the how unstable and fragile the Weimar Germany was, which eventually led the nation to become a Nazi state.

Berlin holds a certain appeal for me. In my opinion, it exudes freedom, youth, and wildness. It's like Bob Dylan as a city. The CCC (Chaos Computer Club), a hacker association I've heard tales about, was founded in Berlin, and it has attracted me to go there and experience the creativity of Berlin hackers.

"Victoria," a low-budget film winner of the 2015 Berlinale Best Film Awards, tells the story of a Spanish girl encountering a group of German boys and becoming a part of a bizarre midnight adventure on the streets of Berlin. It remains one of my top five movies of all time, especially for the original film soundtrack composed by Nils Frahm. My first viewing was at one of the HKIFF's screenings. The immersive experience created by the cinematography and the soundtrack was mesmerizing. The entire film was captured in a single take (or perhaps two? but you wouldn't notice it), taking viewers on a journey through Berlin streets unfiltered, as if you are a flâneur, right there with them.

I have two friends who used to live in Berlin: one is an architect, one is a software engineer. They both have unique characteristics, styles, and interesting backstories.

I haven't been to Berlin yet, but these memories make me crave a visit. My intention is subjective, yes; admittedly, I haven’t taken many other factors about the city into account—neither good nor bad. But I delight in imagining how it would be, and if experiencing it might decharm my fantasy, that's fine.

"Babylon Berlin" was refreshing and provided historical context unfamiliar to me. During that class, I made a visual map of the characters' relationships on Miro, which I found quite helpful for understanding the plot. Here it is:

This will be more like a brief recap of my past few years. A few important dates:

• Ten years ago, I made the rather rash decision to work as a software developer instead of going to college.
• Eight years ago, I made a decision of starting a tech company with my friend, instead of going to college, again.
• Almost three years ago, I left my first startup, and was finally coming back to college to finish my undergraduate degree.

Let's just start with the most recent one, which happened during the COVID period.

College Application, 8 years after High School

To be honest, when I reapplied to university in 2020, I wasn't sure if any school would admit me. Back in 2012, I applied to a few U.S. schools and received offers from several of my dream schools. But in 2020, I was no longer the fresh high school graduate; I was older than the typical applicant, and my outdated standardized test scores didn't quite reflect who I was then. Plus, I was up against a pool of talented high schoolers, where the number of applicants was much greater than in my time. So, filling out those 20 applications while juggling my duties as a startup founder was exhausting.

Of all the schools I applied to, I had only one video interview, quite a pleasuring one. An email hit my inbox one day from the MIT admission office, I was surprised, and thought it was spam. But no, it was an email from an alumni volunteer saying she would be doing an interview with me, her name was Susan. With a few days to prepare, we did the interview via Zoom. There is one moment I now recall quite clearly.

"Hello Tommy – oh hey, you are wearing the shirt!"

Susan said at the beginning, pointing to my somewhat faded shirt, which has the equation $\frac{E}{C^2} \sqrt{-1} \frac{PV}{nR}$ printed on the front – essentially 'MIT' in a nerdy form. I bought this shirt while traveling to Boston in 2016, a pivotal moment when I decided to continue with my startup instead of pursuing a degree again. Meeting Prof. Gilbert Strang of MIT was a special moment; he inspired me to explore MOOCs back in the summer of 2011, when I was part of a team translating his Linear Algebra course from English to Chinese. He taught me that teaching and learning are lifelong tasks. His spirit of perseverance, which has sustained him for over 50 years of teaching – and still going strong on YouTube, beloved by millions – deeply impacted me. I'm grateful for his encouragement and trust, and for his recommendations, offered twice, even eight years after our first online meeting via email. Susan's greeting brought back these memories back. I had been so determined to study at MIT, influenced by Prof. Strang, the hacker culture, the city's walkability, and the wonderful people I met there. I was rejected twice. But who knows – perseverance is the key – according to one of MIT CSAIL's posts.

The other day in February, when I was working from home, I received an email from the UIUC admission office. From that moment, a bond was formed between me and this place – a place I never learned much about, a surprise to my life. So, out of 20 schools, 19 rejected me. I was about to embark on a new journey in the cornfield.

There are two main reasons that pushed me to come to the U.S., one is that my girlfriend is pursuing her doctoral study here, and we don't want to separate for a long-distant relationship. The other one is that, deep down in my heart, I wanted to make up for the rash decisions I made. The degree may just be a piece of paper, but its value largely depends on the student's proactivity during these two to four years. One could just pass all classes and then graduate, while another might do more – choose to engage more deeply, connecting with professors, peers, and various communities. The network created, not just the friendships or connections but also the advantages provided by a formal institution, could be a significant step forward in their learning.

Every life lesson comes to a cost, whether it's about money time, or both. Embrace it and move forward.

UIUC

I have just concluded my fifth semester here at UIUC, and so far everything has gone well. I'm going to quote some of the questions that my course advisor asked me a year ago to describe my up-to-date experience.

Are you happy with your grades? – a faculty mentor

Do I have a good GPA? Sure, I am satisfied with any grade above $\texttt{3.5}$. But more importantly, did I meet some interesting faces? Yes, in general, CS @UIUC is a resemblance to MIT in terms of the hacker culture, in my humble opinion. I've had a great time learning new things about the campus, the professors, and the students here.

Did I learn anything? Well, for the first two years, most of the CS technical core wasn't too challenging for me, as I had more experience than many of my peers, which is unfair to say. However, I gained a lot from picking up the maths. I felt really positive about revisiting my knowledge in most of the math courses at UIUC, especially with Prof. Matthew Russell in Calculus III and Prof. Hongye Liu in Probability & Statistics for CS. Not surprisingly, I made the most comprehensive notes in these two classes as well.

On another note, what I am really amazed about is how well CS @UIUC is doing in computing education, and it's still improving. Take CS 124 for example, the equivalent to Harvard's CS 50 (fun fact, Prof. Malan of CS 50 is a friend of Prof. Colleen, one of CS 124's professors). This course is the starting point for any student new to CS, and its structure is thoughtfully designed to suit a wide range of student demographics. Prof. Challen and his team developed an interactive system for students to code around inside a web browser of their own programming language choice, with different versions of interpretations for certain objectives. This project later has a neat domain called learncs.online. It has the most structured and intuitive representation for learning in a web page setup I've ever seen. Compared to what I researched when I was doing my EdTech startup and teaching students in Hong Kong K-12 schools, I would say CS 124 can be a generous and fulfilling experience for programming novice.

There are other people and groups in the department doing relevant things. Prof. Chad Lane is one of the professors I approached when I entered UIUC and seeking research opportunities. He later introduced me to Prof. Mike Tissenbaum, where I spent almost all of my UIUC time at his lab, Pixel Playground Lab. There, I collaborated with his PhD student, Casey Smith, on the HCI & EdTech project, REACH, which aimed to improve remote maker space learning through a projector-camera system. I once imagined this could be an alternative or pilot version of Bret Victor's Dynamicland. We are still working on making this setup accessible to all audiences. Another influential faculty member I met was Prof. Geoffrey Herman of CS 233 (Computer Architecture with MIPS), is conducting many research on improving student's performance in CS classes. The PrairieLearn is one open-source learning platform he and his team designed to digitize math and CS learning, for a better experience for both teachers and students. While some may find its usability challenging for certain courses, its modular design shows potential to replace traditional learning management systems in the future. As of now, I'm seeing more than five U.S. universities are using it, at least according to the institution login options I saw today. Prof. Angrave of CS 341 and his team are doing research and development on the accessibility for online learning, called ClassTranscribe, while also providing an in-browser VM for students to learn system programming. Dr. Ziang Xiao is a friend I met through a local film development lab, worked on a project at the university focusing on evaluating and improving engineering students' spatial visualization abilities across various demographics. There are just many more interesting people I can name if I have more spaces here.

To elaborate, there's an additional factor that motivated me to pursue a degree: gaining more exposure to fundamental CS research in an academic setting. Fields like computer vision and graphics, compilers, have always fascinated me since high school. Projects from the MIT Media Lab, such as the Sixth Sense project, largely influenced my love for Computer Science and its intersection with other creative or technical fields.

Do you know what you want to do after you graduate? Do you have any questions about research and PhD programs? – a faculty mentor

When I first saw these two questions, I pondered for a while. My simplest answer is a straight inequality: 

$$P(\text{Applying for PhD}) > P(\text{Finding a job}) > P(\text{A new startup})$$

At this point, I can either start fresh (first option) or start from scratch (latter two options). As an outlier, I will have to weave my own path from here. 

PhD may be a good option for me, simply because I want to go in-depth into specific fields. During my eight years in the industry (2013-2021), I had the pleasure of working with people across various fields and sectors: software, hardware, supply chain, design, architecture, photography, education, and VCs. I was in Beijing, Shanghai, and Shenzhen, cities with opportunities. This period of experience has molded my mindset to think in a multidisciplinary way. While breadth is valuable, I also want to try my hand at fundamental CS research. In the context of when AGI becomes true in an (if) not distant future, it can definitely be good at everything, but it's not going to excel at everything (otherwise humans are doomed; yes, I'm that pessimistic). 

There's a good chance that I might get rejected from every program; I always consider the worst-case scenario when making decisions. For now, I'm choosing to ignore this possible outcome. If the worst actually happens, well, it's another story to tell. Prof. Jeff Erikson's tale tells me that anything is possible.

Studying at UIUC is something I never thought about, it is a fortune to be here meeting all kinds of people around campus. I will be always thankful to that admission committee for their decision. I will be paying off my college tuition myself for undergraduate, and I think this is going to be the proudest thing I have done during these four years.

Time in the U.S.

This post should not only be just storytelling. It should be more lively. I should write something about my spare time here, which is mostly spending time driving around with my loved one. Since I got my driving license in 2022, we started enjoying road trips and exploring new cities we've never been apart from the Bay Area, traveling North to Sonoma, Napa, Davis, Klamath, Portland; and East/South to Yosemite, Lake Tahoe, Santa Cruz, Monterey, Death Valley, Las Vegas, Los Angeles, Pasadena, Irving, San Diego. It was definitely a joy.

Other than the road trip, we were just working and studying. Oh, right, both of us started to learn a new foreign language in 2023. This happened when I took a German culture class in 2023 and started learning German; and she started to pick up French when she first learned during her undergraduate period. Despite I only spent about 10 minutes on average on Duolingo, it made me keeps the desire to learn this language. Time to take a German language class next semester.

There's Always a New Chapter

Each year, as we complete another orbit around the sun, it marks a special moment. We set new goals, carry forward some from the past, and celebrate those we've achieved. I certainly reached a few milestones in 2023 and am carrying others into 2024. Every year is a new chapter in my life, yet each one is deeply interconnected. One goal I'm setting for myself is to write more posts here in 2024, whether to consolidate my thoughts and ideas, or just to share photos, or just QotDs.

Here is to 2024: stay kind, work hard, enjoy spare time, as always. Like GD, like ADAM, be optimistically optimizable.

In Chinese, there is a saying that suggests going for a hike on the first day of the year to help you climb higher and higher throughout the year. This saying is known as "新年登高, 步步高升," which symbolizes taking steps that lead to continuous improvement. This year, we followed this tradition and went to Mt. Tamalpais, a place we consider the best spot to overlook the bay and relax.

A friend then took us to Bernal Heights, another fantastic spot to overlook the bay area, but we had a better view of the East Bay.

Indeed, the Ricoh GR II is a camera that does the best at helping me to capture the best moment of my daily life – and the world I have experienced – throughout the past six years. It has the most petite body compared to other competitive models (e.g., Sony RX100) while maintaining the best image resolution of its kind. It is undoubtedly the best APS-C camera out there on the market. Among all the upsides of this model, I specifically love the macro effect, as well as the unique “Daido” (high-contrast black and white) preset.

I bought my GR II in Hong Kong around 2016, and I’ve been bringing it with me as much as possible since then. In fact, I think I’ve used it for every possible ways. Street candid, architecture, landscape, portrait, macro, product… I’d recommend you buy one if you are new to and interested in film and digital photography, as well as want to bring it with you every day.

Here are some photos I took in the past few years:

Now back to the topic. Last year around December, I noticed the shutter curtain didn’t open (see picture below) when I powered it on. But it functioned well after taking one photo, and the process usually took quite a long time (due to the all-black scene with the curtain, not opening). So I googled for some answers. Luckily, I shared the same issue that Michael from StreePhotoTip encountered – and probably every Ricoh GR II user around me has the same issue if they used it for a long time. Michael assumes that the issue is due to the long-term use of the camera. To be precise, the lens cable was broken after frequent stretching when the main board controls the lens to push it out and in.

Back in Shenzhen, I could easily find a place to repair it without hassles. On Taobao.com (the Chinese equivalent of eBay), there are tons of repairing services I could contact to fix my GR. I know it because the last time the LCD screen was broken, it cost me roughly 200 RMB (around USD 30) to replace the parts, and worked perfectly. So yeah, I missed the good old days when I could visit Huaqiangbei, find electronic parts easily, and get them at the lowest price (and with decent quality) possible.

While I was feeling nostalgic, I knew the repair price here in the States must be pricey. After I googled the camera repairing service for a while, some of them either have bad reviews or the costs are higher than expected. However, Ricoh US provides an official repairing service but sending the camera to the repairing site costs around USD 250. I’m not going to waste that much money on an old camera.

Eventually, I went to AliExpress, an international version of Taobao.com, and found a store that sells the flex PCB I wanted to replace manually. Here is what it looks like:

The most satisfying thing is that it only costs me 14 dollars, including the shipping fee. The shipping period isn’t that long; it arrives at my door after about two weeks.

Tearing down the camera

Then, it was time to tear it apart. Here are two videos I found on YouTube. I truly appreciate the author’s efforts in making these content public.

• Ricoh GR (II) Disassembly: Instructions on how to dissemble the camera.
• Ricoh GR (II) Lens Housing Disassembly: Detailed instructions on dissembling the lens module.

It was pretty easy at first. The level of modularization of the camera’s mechanical structure exceeds my expectation – I could easily open the case with the standard screwdrivers and tweezers.

And here’s what it looks like when I tore all the necessary parts apart:

The hardest part is dissembling the lens module. I struggled with the process for a whole night – especially the parts where the stepper motor controls the curtain.

Now I need to pull off the old and broken PCB and replace it with the new one:

Above is the old flex PCB. As you can see, the apparent fracture in the middle of the flex PCB is the problem.

Among all the parts inside GR II, the CMOS sensor board is my fav one. I appreciate how much effort Ricoh’s engineers put into this compact body while maintaining the elegancy and precision of it all.

Putting everything back together

Because the front side of the lens module is broken, I could not add the front cap back. So instead, I installed the GH-3 lens hood and had the lens not exposed to the air. I call it, Ricoh GR II – Cyberpunk edition.

It was the Autumn of 2016. I finally met Prof. Gilbert Strang and audited his lecture at MIT campus. He was just affable and witty as he was in his lecture series. We briefly discussed my learning experience in China when I was in high school and the progress of the Chinese translated version of his book Introduction to Linear Algebra. This unforgettable rendezvous reminds me of my early days in high school, when I was a part-time translator of MOOC lectures, which had not yet become a phenomenon in China. I was responsible for adding bilingual subtitles to Prof. Strang’s Introduction to Linear Algebra lecture series. The experience of making high-quality MOOC resources accessible to fellow students in China was indeed a milestone for me. Self-learning and home-schooling became the ways I engage with the world and knowledge. The absence of a formal college education does not seem to extinguish my curiosity to learn but spark my interest to keep learning through practicing.

In high school, I did two things that have had profound impacts on me. During my first semester of 10th grade, I gathered a group of computer enthusiasts and started the only engineering club in our high school. As the founder, I wanted to create a platform where students could engage in hands-on programming and engineering of LEGO RCX/NXT robots using the LEGO Mindstorms NXT software and LeJOS. I also held regular sharing sessions with club members, introducing the (then) latest development of maker projects, which of course included many MIT Media Lab projects. This club marked the starting point of my maker life and motivated me to invite more like-minded high school students from all over the country to form a community of aspiring computer engineers. Later in my 11th grade, I founded the nation-wide Adolescent Developer Community (ADC), which is the first of its kind in China. I devoted much of my spare time to organizing the annual ADC conference, including inviting keynote speakers both from academia and industry as well as securing sponsorship from tech giants.

To date, the engineering club I founded is still a nice makerspace for young-generation students to learn the fundamentals of computer science and engineering. The ADC annual conference was later held for three years consecutively. I harvested friendship with like-minded students as well as developed budding leadership and entrepreneurship in the STEM field.

In the summer of 2013, when I received offers from several U.S. universities, a software engineer, Kenny, invited me to work with him at Opera Software. I was encouraged by his warm, tempting invitation and generous recognition of my skill set. If working with a group of brilliant engineers would be my future career, why would I wait till four years later? Why would I not pursue my dream now? At that time, I believed that computer science would be ultimately a skill-oriented subject and I should begin my journey on learning through practicing. I eventually made a decision of passing the ticket to college and working with Kenny at Opera Software. Admittedly, my perspective on college education might be naive and biased. But I was determined to pursue a career in the industry I love, which was just a different path for my future. I enjoyed my time at Opera Software and felt satisfactory with the projects I was involved in, which ranged from developing and testing Sphinx HTML5 Game Engine with the Game Engine team and building the developer relationship community for this project. Thanks to Fredrik and Kenny, I was able to learn from the best. After I left Opera, I focused my career on the field of product design. Later I set up my own design studio, which provided design services to many startup clients. In a word, my post high-school experience was nothing but rewarding and important to my decision of returning to school.

In 2016, I met Tony, a PhD student at the Chinese University of Hong Kong. We shared common interests and visions in STEM education and we were both deeply inspired by such technological pioneers as Nicholas Negroponte and Marvin Minsky. The two of us just clicked immediately. I had acquired sufficient skill set from previous work experience and still entertained the idea of advancing the field of STEM education, which was a seed planted when I was in high school. Later Tony and I co-founded the ed-tech startup “CocoRobo”, a company that provides a one-stop solution for K-12 STEAM education.

Our mission is to create easy-to-use products that embody the concept of PBL curriculum for teachers and students in K-12 classrooms. Our R&D team developed a series of ed-tech products over the span of four years. The hardware part contains a set of plug-and-play electronic modules with B2B connector. There are currently two generations. The first generation focuses on STEAM education solution. The second generation centers on AI STEM education solutions, which consists of an IoT module and an AI module. The software platforms include CocoBlockly—an online visual programming environment, CocoCloud—a cloud platform for IoT data collection and visualization, and CocoEducation—an online education platform for our users. We also implemented aforesaid core technologies into a series of theme-based toolkits surrounding such topics as Creative Electronics, Robots & Mechanicals, and Human-computer Interaction, which have huge appeal to kids. In this year’s annual meeting of PyCon China, I shared with the audience my experience of implementing Python and MicroPython in K-12 STEM classrooms in the past four years.

Reflecting on the past four years of starting my own business in the field I am passionate about, I find myself establishing the skills beyond my imagination. Surely, my leadership and entrepreneurship transcended from a club founder to a conference organizer and then to a CTO of a company. Still, I have much to learn and achieve as a leader who is seeking innovative technology solutions for the world. This year, our company participated in a project called “Future Classroom.” We visited a secondary school in Enshi, Hubei province, which was one of the hardest COVID-hit regions in China. We provided our intuitive products to kids in Enshi and helped them to use our products to learn computer programming and the concept of design thinking. The whole idea of Future Classroom, or making education accessible to kids in every corner of the world, reminds me of my old days in high school. In Prof. Strang’s class and the engineering club, I benefited from approachable education and tried to put ideas into practice. I cannot help but ask myself what else I can do in the future.

My life has progressed from Introduction to Linear Algebra, yet I can trace my path back to that inspiring course I found online.