Projects

Regime-Switching Factor ETF Allocator
Technologies: Python, HMM, Random Forest, cuML (GPU), PyTorch, Streamlit
GitHub Repository

Production-ready regime-switching factor ETF allocator using Hidden Markov Models (HMM), supervised learning, and threshold baselines
Achieved 13.1% CAGR with 1.18 Sharpe ratio using GPU-accelerated supervised learning model
Built with walk-forward testing, realistic transaction costs, GPU acceleration, and bootstrap uncertainty quantification
Implemented three regime detection methods: threshold rules, Gaussian HMM, and Random Forest (GPU-accelerated)
Features include regime-conditioned portfolio allocation, monthly rebalancing, and comprehensive backtesting framework

PaperPal: AI-Powered Research Assistant
Technologies: Go, Gin, gRPC, FAISS, LangChain, Next.js

Engineered academic QA backend, doubling PDF throughput and cutting latency by 40%
Implemented LangChain RAG agent with 85% top-3 recall and Next.js frontend

Insider Threat Detection
Technologies: SeqGAN, LSTM, Multi-Head Attention, PyTorch, PostgreSQL

Enabled proactive insider-threat detection with SeqGAN+PostgreSQL, rebalancing 3 threat classes
Trained LSTM with multi-head attention for sequential classification, achieving 80% accuracy