Skilled in Full-stack engineering, AI engineering. Interested in machine learning research
Most recent work is the Kimi.ai, received angel fundraisings. Live on App Store. Did 95% of engineering of the beta version for free.
Conducted recruiting for Imaginix: github.com/imaginix-inc/react-interview/branches
Education
Took CS256 topics in artificial intelligence and CS257 topics in machine learning besides computer science.
Courses taken besides computer science: statistics, statistical inference, linear regression, optimization, machine learning, speech and natural language processing, deep learning, optimization in machine learning.
Work Experience
• Built a full stack live2d AI Anime voice chatting app using React Native, Next.js, Supabase, tRPC, and Tamagui.
• Implemented text and audio streaming in React Native using fetch polyfill, with OpenAI and Elevenlabs.
• Built live2d animation using both Unity and Pixi.js. Implemented motions, background music, and speech.
• Oversee the entire workflow from design to product. Led a team of 4 developers. Assigning issues including
Feb 2023 - Apr 2023
• Implemented Dolby Vision PC ICM generation SDK with .NET Core and support portal with Blazor.
• Improved the UI/UX & API Design for Dolby Vision support portal with ASP.NET, jQuery, and Vue.js.
• Finished twice as many features/issues as my team anticipated before my onboarding.
May 2022 - Aug 2022
• Built OpenShift Operators with Operator SDK in Golang which impacted >50 user issues and test cases.
• Built OpenShift Console Plugins with React and TypeScript that impacted two major releases.
• Built department-wise collaborative dashboard and editor in React and SQLite impacted >200 people.
• Involved in quality assurance of the OpenShift console, focusing on automating test cases with Cypress.
• Contributed to OpenShift related projects; responsible for debugging Jenkins CI and fixing Ruby scripts.
Feb 2022 - May 2022
• Built a C++ library based on HDF5 that reads and parses massive volumes of hierarchical data in parallel.
• Developed parallel CCA algorithm in C++ and NumPy; also, gradient descent algorithm in NumPy.
• Distributed the NumPy implementation workload using the Dask library and the C++ with MPI.
• Implemented feature engineering and data augmentation for videos based on pix2pix and TensorFlow.
• Participated in model training & serving with TensorFlow Serving; exposed APIs for the RSD to consume.
Jun 2021 - Aug 2021
- Built a landing page, article page, social page, admin page, and rich-text article editing/publishing page with React, Node, and MongoDB.
- Integrated the admin article publishing page with Notion, so my client was able to sync between his Notion documents and the admin page seemlesssly.
Jul 2021 - Aug 2021
- Responsible for optimize the data tracking within a video playing component for a research project of Prof. Zhang. (sme.cuhk.edu.cn/en/teacher/182).
- Responsible for optimize performance for a image heavy website for data tracking. Responsible for compressing assets on the server.
Jan 2021 - Feb 2021
• Implemented an interactive heatmap component with D3.js and Vue.js; built a desktop app with Electron to boost clients' satisfaction.
• Implemented a C++ library on Raspberry Pi 4 to to handle general-purpose input and output (GPIO) for the company's sensor.
Jun 2020 - Aug 2020
• Implemented and deployed web scrapers with Puppeteer; set up Chron jobs and Google Sheets on GCP.
• Developed Puppeteer extensions and scripts to circumvent bot detection mechanisms.
Side Projects
- Built an interactive law answering and document template generation with GPT-4 & RAG on pgvector (HNSW).
- Built the front-end UI and backend streaming and querying with React, TypeScript deployed on Vercel.
- Scraped multiple states law data, splitted and implemented Neo4j graph structures for better retrieval.
A React Native iPad LeetCode client. The project was sold to an anonymous startup. Only an early stage of the project is kept.
Projects
Wrote the frontend, intermediate, and backend of the compiler in Java.
- Wrote the MIPS assembler and simulator in C++ for instruction execution. Kind of like the backend executor + frontend/intermediate stuffs that I did for the compiler course at SJSU.
- Wrote a single-cycle and a multi-cycle pipelined CPU in Verilog with hazard handling.
• Implemented traditional object detection algorithms with OpenCV with manual feature selection.
• Trained & fine-tuned STOA models including Faster R-CNN, SSD, and YOLO to compare performances.
• Wrote Bash scripts to train and log results after individually setting up cloud GPU resources.
(The repo was private due to the setting of GitHub Classroom)
• Reproduced a RoBERTa ChatGPT/Human text classification model from a paper achieving >0.999 acc.
• Set up multiple LLM APIs and on-device models to produce more training text which achieved better
generalizing performance over the baseline model.
(The repo was private due to academic requirements and the setting of GitHub Classroom).
For problems including Parallel-Odd-Even-Transposition-Sort, Mandelbrot Set Computation, N Body Computation, and Heat Distributions, I implemented different parallel programming algorithms with Pthreads, MPI, and OpenMP and ran benchmarks and tests on all of the libraries.
• Implemented data preprocessing with NLP techniques like tokenizing, normalizing, and padding.
• Iterated pre-trained models like BERT, DeBERTa, and RoBERTa and parameters and got a silver medal.
• Implemented components OS including process in kernel, threading with lock, memory virt, IO, and fs.
• Implemented algorithms like deadlock avoidance, resource alloc, CPU scheduling, and page replacement.