ALICE – Adversarial Co-Evolutionary RL Training Environment
Built an adversarial co-evolutionary reinforcement learning environment using PyTorch, TRL, GRPO, Unsloth, Qwen2.5, and Gradio. Implemented the OHRV cycle in which an Oracle generates hard scenarios, a Hunter attempts them, Repair corrects failed trajectories, and Verify validates them. Added auto-scaling environment difficulty, custom reward shaping, GRPO training, and a real-time Gradio dashboard for reward curves, win-rate evolution, and difficulty progression. Selected among the top teams in the Meta PyTorch OpenEnv Hackathon Grand Finale out of 52,000+ developers.
ORACLE – Autonomous AI Trading Agent
Built an autonomous prediction market agent for the AI Forecasting Hackathon 2026 at the University of Chicago using ensemble forecasting, news sentiment, market signals, Kelly Criterion position sizing, and the Kalshi API. Engineered real-time monitoring to calculate gaps between estimated edge and implied probability and execute trades above a calibrated confidence threshold.
Hybrid RAG Travel Planner
Built a FastAPI backend combining Neo4j knowledge graph traversal with ChromaDB and Pinecone vector search, achieving a reported 92% improvement in personalization. Orchestrated multi-step LLM reasoning with LangGraph to resolve complex multi-hop queries at sub-second latency.