Tchelp.org — Tech Support Platform
Full-Stack
Founder & Lead Developer · 2025 — Present
- React + FastAPI, Supabase, Cloudflare CDN.
- Launched to 1,000+ unique visitors and 13,000+ requests in the first month.
Visit site
YOLO + PPO Game Agent (Vision → Policy)
AI/ML
PyTorch · Ultralytics YOLO · Reinforcement Learning
- Trained a vision-based agent: YOLO detections are encoded into a 26-D state and fed to a PPO policy.
- Used behavioral cloning from a heuristic expert to bootstrap PPO and prevent degenerate "camping" behaviors.
- Implemented a reproducible pipeline:
GameCore (training) + web_app (Socket.IO demo).
Visit repo
Logical Benchmark — Research
Research
LLM strategic reasoning on rule-based games (Advisor: Prof. Tony Dear)
- Python framework for self-play and evaluation (win rate, move validity).
- Designed metrics to assess model decision-making.
AI-Powered Reading Feed
Full-Stack
GraphQL middleware + AWS EC2 backend
- Personalized recommendation pipeline and scalable endpoints.
Visit repos
ImposterAI — AI Social Deduction Game
Hackathon
MIT Hackathon · Raspberry Pi · Arduino · AI
- AI-powered Imposter-style tabletop game where AI acts as both host and deceptive player.
- Physical lotus-shaped device with LED coasters; Raspberry Pi + Arduino for hardware control.
- AI generates context-aware lies and deception strategies during live gameplay.
Visit repo