I am an independent researcher and engineer working at the intersection of reinforcement learning, quantitative finance, and systems engineering. This project is the culmination of years of studying why AI trading systems fail — and building one that doesn't cut corners.

Why This Project Exists

Most "AI trading" products fall into one of two traps: they are either overfitted backtests masquerading as strategies, or black-box SaaS tools that hide their methodology. I wanted to build something different:

• Open about methods: Every experiment is logged, every metric is on held-out data, and every claim is caveated.
• Realistic about costs: Indian markets have material transaction costs per leg. Ignoring this destroys alpha. We model every fee, tax, and slippage component.
• Rigorous about validation: Walk-forward testing, adversarial stress tests, and paper trading gates before any real capital touches the market.
• Honest about stage: We are in research. One published result. More in pipeline. No live capital yet. No guarantees.

Background

My work spans machine learning research (deep RL, representation learning), quantitative systems (backtesting, risk modeling, execution simulation), and production software engineering (distributed systems, performance optimization). I have spent the last several years focused specifically on the challenges of applying RL to financial time series — a domain where standard assumptions break down quickly.

Principles

• Validation over vanity metrics: A Sharpe of 2.0 on training data means nothing if test Sharpe is 0.3.
• Costs are not optional: Any backtest without realistic transaction costs is fiction.
• Risk is part of the objective: Drawdown constraints, concentration limits, and tail-risk awareness must be baked into the agent's reward function, not applied as afterthoughts.
• Speed enables intuition: If an experiment takes days, you cannot iterate. GPU acceleration and vectorized environments are not luxuries — they are requirements.

What I'm Looking For

This is a research-stage project seeking the right collaborators, not the most collaborators. I am particularly interested in:

• Investors who understand the difference between a validated prototype and a live product
• Quant researchers working on regime adaptation, meta-learning, or cross-sectional modeling
• ML engineers who have shipped RL systems in production
• FinTech teams who need an Indian-equity RL engine they can run in-house