Portfolio
Filter
Showing all 21 projects
-
WashU Chinese Geeks Club
Website
Club Sep 2024 - Present- Serve as an event planner for a community of more than 300 WashU Chinese students interested in computer science and career development.
- Organized and participated in non-hackathon events covering LeetCode practice, interview skills, course selection, and academic planning.
- Supported a course helper program that assisted 15 students with schedule planning.
- Situation
- WashU Chinese students needed a supportive community for sharing computer science interests, career experience, and academic guidance.
- Task
- As an event planner, I was responsible for organizing learning and community events around algorithms, interviews, and course planning.
- Action
- I planned and participated in the club's non-hackathon events, including LeetCode practice and course-selection sessions, and supported the course helper program.
- Result
- The club grew to more than 300 members, and its course helper program assisted 15 students with schedule planning.
Event planner for a 300+ member WashU community supporting Chinese students through computer science, interview, and academic-planning events.
Skills: Event Planning, Communication, Organization, Leadership
-
circtime
GitHub
Infrastructure Monitoring, Web Application Jul 2026 - Present- Extended a React and TypeScript uptime dashboard into an interactive 3D radar for visualizing infrastructure, services, and network dependencies.
- Implemented configuration-driven URL and TCP monitoring, degraded-state detection, dependency analysis, DNS topology discovery, and up to one year of uptime history.
- Automated checks every 15 minutes and static deployment through GitHub Actions and GitHub Pages, currently mapping 67 configured infrastructure, service, and network nodes without a separate backend.
- Situation
- My hybrid-cloud environment spans VPS hosts, a NAS, GPU infrastructure, DNS and routing layers, and many self-hosted services that need to be inspected as one system.
- Task
- Create a reusable, configuration-driven dashboard that monitors the infrastructure and communicates service health, history, and dependency relationships without a dedicated monitoring backend.
- Action
- I extended the initial React and TypeScript dashboard with a Three.js concentric 3D visualization, a YAML configuration pipeline, URL and TCP probes, degraded-state thresholds, dependency diagnosis, DNS discovery tools, administrative debugging, and a GitHub Actions monitoring workflow.
- Result
- The dashboard maps 67 configured nodes, checks available endpoints every 15 minutes, retains up to 365 days of history, and deploys as a static GitHub Pages application configured through one YAML file.
A configuration-driven 3D uptime radar that maps 67 infrastructure, service, and network nodes while monitoring available endpoints through GitHub Actions and GitHub Pages.
Skills: React, TypeScript, Three.js, GitHub Actions, Infrastructure Monitoring, DNS, DevOps
-
ImGenHCT
GitHub
Computer Vision, Image Generation Nov 2025 - Dec 2025- Served as the final project for CSE 5519: Advances in Computer Vision.
- Explored highly compressed image tokenization combined with text-to-image generation, using a TiTok-style 1D tokenizer to encode images into minimal latent representations.
- Built an encoder-decoder pipeline that uses CLIP and VQA metrics to evaluate semantic alignment between generated images and text prompts.
- Trained a token-to-token model on 400,000 samples for 35 epochs; it reached a 0.79 VQA score over 50 prompts with five seeds each, compared with 0.75 for VQA-based optimization.
- Situation
- Diffusion models provide strong image quality but require iterative denoising, while one-step generators often sacrifice fidelity.
- Task
- Investigate whether highly compressed 1D TiTok image tokens could support efficient one-step text-to-image generation.
- Action
- I co-led the project, designed blocks of the token-to-token model, reviewed code, tested test-time generation on home computers, created visualizations, and co-wrote the paper.
- Result
- After training on 400,000 samples for 35 epochs, the model reached a 0.79 VQA score across 50 prompts with five seeds each, outperforming VQA optimization at 0.75 while generating images in one step.
A one-step text-to-image pipeline that maps text directly into compressed 1D TiTok image tokens and reached a 0.79 VQA score.
Skills: Computer Vision, Image Generation, TiTok, CLIP, PyTorch, Python
-
ATTPDRL
GitHub
Deep Reinforcement Learning, Autonomous Vehicles Nov 2025 - Dec 2025- Served as the final project for CSE 5100: Deep Reinforcement Learning.
- Modeled trailer-truck steering in a Gymnasium environment using Shapely and Pygame, and compared DQN, PPO, and DDPG in trajectory-free scenarios.
- DQN converged after about 400,000 steps for forward driving and 800,000 steps for backward driving, but failed to solve the 90-degree steering task after 5 million steps.
- Situation
- Trailer trucks have nonlinear articulated dynamics that make steering and parking in constrained spaces substantially harder than controlling standard cars.
- Task
- Design a trajectory-free simulation environment and evaluate whether deep reinforcement learning could steer a trailer truck while avoiding collisions and jackknifing.
- Action
- I modeled the vehicle kinematics, implemented the environment with Gymnasium, Shapely, and Pygame, compared DQN, PPO, and DDPG, and ran reward-function ablations.
- Result
- DQN converged around 400,000 steps for forward driving and 800,000 steps for backward driving, but the 90-degree steering task still failed after 5 million steps, showing the limits of the trajectory-free reward design.
A trailer-truck reinforcement-learning simulator where DQN learned straight forward and backward control but exposed the limits of trajectory-free steering rewards.
Skills: Deep Reinforcement Learning, PyTorch, DQN, PPO, DDPG, Python
-
IMSB personality test site
Frontend on GitHub
Web Application Apr 2026 - Apr 2026- Built a prank personality-test site that returns random matching rates rather than performing personality analysis.
- Completed the site in four hours and promoted it with a streaming-platform video, attracting 2,000 test takers in seven days.
- Situation
- A prank personality-test concept needed a complete web experience that looked credible while intentionally returning random matches.
- Task
- Build and promote the full site within a very short development window.
- Action
- I completed the site in four hours and used a video on a streaming platform to advertise it.
- Result
- The site attracted 2,000 test takers during its first seven days.
A prank personality-test site built in four hours that attracted 2,000 test takers in one week.
Skills: Front-End Development, Back-End Web Development, Databases, Docker
-
SophiasPath
Frontend on GitHub Backend on GitHub
Web Application Sep 2024 - Jan 2025- Built an interactive website for delivering stories and content about philosophers.
- Built a Django and Django REST Framework backend with an administrative dashboard for content management.
- Completed and deployed the full-stack service with Docker and reverse-proxy tooling; feature updates ended when the content provider stopped supplying updates.
- Situation
- A content provider needed an interactive site for publishing stories and educational material about philosophers.
- Task
- Deliver the full-stack application and provide an administrative workflow for managing its content.
- Action
- I built the backend with Django and Django REST Framework, added an admin dashboard, and deployed the service with Docker and reverse-proxy tooling.
- Result
- The full-stack site was completed and deployed; feature development ended after the content provider stopped supplying updates.
A completed Django-based philosophy storytelling site with a REST backend, admin dashboard, and containerized deployment.
Skills: Front-End Development, Back-End Web Development, Databases, Docker
-
Mon3tr
GitHub
Computer vision Jun 2025 - June 2025- Served as the final project for Spring 2025 CSE 559A Computer Vision at Washington University in St. Louis.
- Combined MASt3R and Smooth-Diffusion to test whether 2D image generation could improve 3D reconstruction.
- Implemented the generation pipeline and bridged gaps between the models, but the generated 3D reconstruction was not yet satisfactory or usable.
- Situation
- The project explored whether a 2D diffusion model could fill missing visual information during image-based 3D reconstruction.
- Task
- Connect MASt3R and Smooth-Diffusion and build the generation stage of the reconstruction pipeline.
- Action
- I focused on coding the generation pipeline and connecting the outputs required to bridge the two models.
- Result
- The pipeline filled part of the integration gap, but the generated 3D model was still unsatisfactory and not usable.
An experimental MASt3R and Smooth-Diffusion pipeline that bridged 2D generation and 3D reconstruction but did not yet produce usable models.
Skills: Computer Vision, MASt3R, Smooth-Diffusion
-
Flow-of-Thought
GitHub
Large Language Models, Chain of Thought, self-RAG May 2024 - May 2024- This is an experimental project as Final project of Fall 2024 CSE 561A Large Language Models course at Washington University in St. Louis.
- Implemented a multi-agent self-RAG system with Chain of Thought and Large Language Models. Based on Graph of Thoughts pipeline.
- Situation
- This course project examined whether multiple language-model agents could improve performance on long tasks.
- Task
- Implement a multi-agent self-RAG workflow based on Chain of Thought and a Graph of Thoughts pipeline.
- Action
- I implemented the multi-agent self-RAG pipeline and tested subagent-based processing for longer tasks.
- Result
- Subagents improved long-task performance, but the improvement came with a cost-versus-accuracy trade-off.
A multi-agent self-RAG experiment showing that subagents can improve long-task accuracy at increased cost.
Skills: Large Language Models, Chain of Thought, self-RAG
-
Odoru
GitHub
Web Application (HackWashU Hackathon 2024) Nov 2024 - Dec 2024- Built a HackWashU prototype that streams a reference dance while capturing the user's movement through a webcam.
- Used real-time pose detection and skeletal-motion similarity to provide a continuously updated dance score.
- Focused on frontend implementation and UI design; the hackathon prototype was not deployed after the event.
- Situation
- A HackWashU team wanted to give dancers real-time visual feedback by comparing webcam movement with a reference video.
- Task
- Build the frontend and interaction design for the dance-companion prototype.
- Action
- I implemented the frontend and UI for video selection, webcam capture, live pose feedback, and continuously updated scoring.
- Result
- The team completed a working hackathon prototype with real-time pose comparison, but did not deploy it after the event.
A hackathon dance companion with a frontend for webcam pose comparison and live movement-similarity scoring.
Skills: Machine Learning, Computer Vision, Web Development
-
Cobalt-4
Colorful number project repo - Cobalt-4
Game Development Jan 2026 - Now- Initiated a 2.5D rhythm-game project built around moving tracks and four-key notes.
- Focused on art and UI design and contributed code reviews.
- Completed an internal prototype that has not yet been published.
- Situation
- The team initiated a 2.5D rhythm game using moving tracks and a four-key note system.
- Task
- Define the project's visual presentation and user interface while reviewing implementation work.
- Action
- I focused on the art and UI design and contributed code reviews during prototype development.
- Result
- The team produced a working internal prototype that has not yet been published.
An unpublished 2.5D rhythm-game prototype for which I designed the art and UI and reviewed code.
Skills: Unity, Digital Illustration, Programming
-
Khaki-51
Colorful number project repo - Khaki-51
Game Development Sep 2023 - Oct 2023- Initiated a new chess game project.
- It works like chess (no it doesn't) but have action points that can do various operations to move the chess.
- Situation
- The project explored a chess-like game in which action points replace conventional movement rules.
- Task
- Initiate a game prototype around action-point-based board operations.
- Action
- I initiated the Unity project and implemented the core action-point movement concept.
- Result
- The project produced an experimental chess variant centered on action-point operations.
A small Unity experiment that reworks chess-like movement around an action-point system.
Skills: Unity, Programming
-
Raven-24
Colorful number project repo - Raven-24
Game Development Mar 2022 - Sep 2022- Designed a puzzle-game concept about a girl using properties of time to save her friend while trapped in an endless cycle.
- Worked as the game designer, but the project did not produce a prototype.
- Learned the importance of bounded planning before beginning implementation.
- Situation
- The project followed a girl using properties of time to save her friend while trapped in an endless cycle.
- Task
- As the game designer, define the time-loop narrative and puzzle-game concept.
- Action
- I designed the game concept and its time-based narrative without first constraining the project to a bounded implementation plan.
- Result
- The project did not produce a prototype and taught me to establish bounded scope and planning before starting implementation.
An unprototyped time-loop puzzle-game design that taught me to bound project scope before implementation.
Skills: Unity, Programming
-
Graphical Applets for Computational Geometry Algorithms
GitHub Live applet
Web Application Dec 2023 - Dec 2023- Developed graphical applets using JavaScript to visualize computational geometry algorithms.
- Implemented an interactive range-tree and segment-tree visualization as the final project for CSE 546: Computational Geometry.
- Situation
- Computational geometry data structures are easier to understand when their spatial operations can be inspected visually.
- Task
- Build an interactive JavaScript visualization as the final project for CSE 546: Computational Geometry.
- Action
- I implemented a graphical applet that demonstrates range-tree and segment-tree behavior in the browser.
- Result
- The completed applet is published as an interactive computational-geometry teaching demo.
An interactive JavaScript applet for visualizing range-tree and segment-tree operations.
Skills: JavaScript, Computational Geometry
-
Applets for Algorithms
Site deployed Leetcode profile
Web Application Dec 2023 - Present- Developed zoo of algorithms used in Competitive Programming contests, especially Leetcode.
- Serve as extension for final project for CSE 546: Computational Geometry.
- Leetcode top 0.64% (updated on Apr 2026), without use of AI.
- Situation
- I wanted to extend a computational-geometry course project into a broader reference for competitive-programming algorithms.
- Task
- Develop and maintain a browser-accessible collection of algorithm applets, especially for LeetCode practice.
- Action
- I expanded the original course work into a deployed collection covering algorithms used in competitive programming.
- Result
- The applet collection remains deployed, while my related LeetCode practice reached the top 0.64% as of April 2026 without AI assistance.
A deployed collection of competitive-programming algorithm applets developed alongside a top-0.64% LeetCode record.
Skills: JavaScript, Advance Algorithms
-
NoteNextra (Notechondria)
GitHub Accessible Website
Full-Stack Application Nov 2023 - Present- Built a personal memory archive and note site with Next.js and the Nextra framework, including LLM-assisted retrieval and utilities.
- Reached about 10 users per week; feature development ended after graduation, with only typo and error corrections continuing.
- Situation
- Personal notes and reference material needed a searchable archive that could support retrieval and lightweight utilities.
- Task
- Build and maintain a web-based memory archive with LLM-assisted retrieval.
- Action
- I implemented the site with Next.js and Nextra, organized the note archive, and added retrieval and personal-utility features.
- Result
- The site reached about 10 users per week; feature development concluded after graduation, and maintenance is now limited to typo and error corrections.
A Next.js and Nextra memory archive with LLM-assisted retrieval that serves about 10 users per week.
Skills: Next.js, Nextra, Web Development, Large Language Models
-
Web Development & Personal Website
Website
Web Development Feb 2020 - Present- Built a website containing learning material in computer science and shared personal interests at www.trance-0.com.
- Created a series of services via docker: git server, cloud storage via NextCloud/OpenList, media server via Immich, CI/CD manager via Jenkins, password manager via Bitwarden, chatbot supports QQ, WeChat, Telegram, and Discord.
- Situation
- Moving daily work across Linux and Windows exposed the cost of depending on one vendor for files, photos, passwords, code, notes, and compute access.
- Task
- Build a portable personal infrastructure layer using open-source services that I could inspect, replace, and operate myself.
- Action
- I deployed and maintained containerized services for storage, photos, passwords, Git hosting, CI/CD, media, chatbots, and remote compute behind DNS and reverse-proxy infrastructure.
- Result
- The platform remains in daily use and taught me to treat reliability as an operational problem involving backups, exposure boundaries, routing, migration, and recovery rather than only application code.
A long-running self-hosted hybrid cloud that replaces proprietary storage, media, password, Git, CI/CD, and messaging services with open-source infrastructure.
Skills: Computer Science, Computer Networks, Web Development, Linux, Docker, Programming
-
Cookistry
Full-Stack Application Jul 2023 - Sep 2023- Initialized project in Insite x ACM x Skandalaris No-Code HackAThon 2023.
- Built a full-stack app for sharing standardized recipes in Python with PostgreSQL using Django Framework.
- Designed the app mechanism and UI.
- Adapted by WashU cooking club for sharing recipes.
- Enabled customized cooking experience with preferred units and accurate time measures.
- Situation
- A hackathon and the WashU cooking community needed a consistent way to share recipes across different units and timing preferences.
- Task
- Design and build a full-stack recipe-sharing application with standardized data and customizable presentation.
- Action
- I initialized the project, built it with Django and PostgreSQL, and designed the application mechanism and UI.
- Result
- WashU's cooking club adopted the application for sharing recipes with preferred units and accurate time measurements.
A Django and PostgreSQL recipe-sharing app adopted by WashU's cooking club for standardized, customizable recipes.
Skills: Full-Stack Development, Django, Programming, PostgreSQL, Databases
-
Using Machine Learning to Analyze Apartheid in South Africa
Data Science Sep 2022 - Nov 2022- Provided technical support in a team using Python, TensorFlow, and GPT-3 to construct models analyzing Apartheid in South Africa and Namibia from the Bohr era to the end of segregation.
- Situation
- A team project applied machine learning to historical material about Apartheid in South Africa and Namibia.
- Task
- Provide technical support for the team's model-construction work.
- Action
- I supported the Python, TensorFlow, and GPT-3 work used to construct the analysis models.
- Result
- I contributed technical implementation support to the team's historical-analysis workflow.
A supporting technical role using Python, TensorFlow, and GPT-3 in a team analysis of Apartheid-era material.
Skills: Programming, TensorFlow, Python
-
TimeBlocks
Colorful number project repo - White-75
Mobile Application Feb 2021 - Feb 2022- Developed an app to help users enhance concentration and relax scientifically by recording concentration time and finding optimal work-relax cycles.
- Features include to-do list management, distraction detection, and personalized reports.
- Built independently using Unity and SQL database on VPS.
- Situation
- The project explored how recorded focus time could help users identify more effective work-and-rest cycles.
- Task
- Build a full-stack concentration tracker with planning, distraction detection, and personalized reporting.
- Action
- I independently implemented the application in Unity with a SQL database hosted on a VPS.
- Result
- The project produced a complete independent full-stack application with focus tracking, to-do management, distraction detection, and reports.
An independently built Unity and SQL focus tracker for measuring work-rest cycles and generating personalized reports.
Skills: Unity, Programming, SQL
-
Gray-75
Colorful number project repo - Gray-75
Game Development Feb 2020 - Feb 2022- Initiated a first-person puzzle-game demo about Alzheimer's disease.
- Led a three-person team, designed scenes, implemented gameplay with Unity and Blender, and supported junior developers.
- Completed a playable demo and video preview.
- Situation
- The project used a first-person puzzle game to explore a narrative about Alzheimer's disease.
- Task
- Lead a three-person team and deliver a playable demonstration of the concept.
- Action
- I led development, designed scenes, implemented gameplay with Unity and Blender, and provided long-term support to junior developers.
- Result
- The team completed a playable demo and a video preview.
A three-person Unity and Blender puzzle-game project about Alzheimer's disease that produced a playable demo and video preview.
Skills: Unity, Blender, Programming
-
Robot Open Autonomous Racing (ROAR)
ROAR Racing training site Fall 2020 results
AI/Robotics Apr 2019 - Nov 2021- Participated in UC Berkeley's Robot Open Autonomous Racing program and worked in a three-person simulation-racing team.
- Manually tuned a PID controller for fast autonomous driving and supported other team members and new participants.
- Placed second in the Fall 2020 ROAR S1 series with a 1,104.00-second total and recorded the fastest single lap at 77.15 seconds.
- Situation
- The ROAR simulation competition challenged teams to control an autonomous car quickly and reliably around a racing environment.
- Task
- Work in a three-person team to improve autonomous driving performance for the Fall 2020 S1 series.
- Action
- I helped manually tune the PID controller, supported team performance work, and provided technical guidance to newer participants.
- Result
- Our team placed second with a 1,104.00-second total and earned the fastest-single-lap result at 77.15 seconds.
A three-person ROAR team using a manually tuned PID controller that placed second and recorded the fastest lap in Fall 2020.
Skills: TensorFlow, Programming, AI