I'm a CS student building evidence-grounded AI agents, developer tools, and data products.
I like projects that go past "call an API": collect real data, retrieve evidence, explain uncertainty, run tests, and ship a usable interface.
| Project | What it shows |
|---|---|
| PatchGuard | Evidence-backed verification for AI-generated Python PRs. It analyzes diffs, runs tests/static/security checks, scores merge risk, and serves a React/FastAPI dashboard. |
| Transfer Stock Analyst | An AI analyst for football transfer rumors, reporter credibility, and listed-club stock reactions, with RAG, local agent runs, scenario simulation, and a static dashboard. |
| 2026 World Cup Predictor | A transparent tournament simulator for the expanded 48-team World Cup, with probabilistic matches, Random Forest support, and a FastAPI demo app. |
| ASL Sign Translator | A real-time sign-language translation prototype using MediaPipe hand landmarks, a GRU sequence model, a WebSocket inference backend, and a Next.js frontend. |
- Search Engine: inverted index pipeline, ranking/search servers, tests, and a small search UI.
- MapReduce: a Python manager/worker MapReduce system with integration tests.
- Full-Stack Social App: Flask, SQLite, React, REST APIs, authentication flows, infinite scroll, and Cypress tests.
- SWMM + Leaflet Infrastructure Map: parses stormwater model files into GeoJSON and visualizes network hotspots on an interactive map.
- DateRank: privacy-first analytics for user-owned dating-app exports, built around local parsing and deterministic metrics.
- Mental Health NLP Retrieval: mental-state classification and personality-aware resource retrieval research with TF-IDF, logistic regression, and BM25 baselines.
- AI agents that cite evidence instead of guessing
- RAG systems over local project data
- CI and verification tools for AI-generated code
- Sports, finance, and market-intelligence dashboards
- ML systems that are honest about uncertainty and evaluation
Python · FastAPI · pytest · Docker · GitHub Actions · React · Next.js · TypeScript · SQLite · pandas · scikit-learn · RAG · LLM agents
- Start with the data contract.
- Make the result inspectable.
- Add tests where the risk is real.
- Keep model outputs grounded in evidence.
- Ship a demo people can actually open.
- Portfolio: kaiwen_portfolio
- GitHub: @KaiwenMo1