Project H — Building an AI-driven options trading engine

What we're building

Project H is a research engine for systematic options trading in the Indian F&O market — Nifty 500 universe, 146 tradeable F&O contracts, fully automated end-to-end.

The system is built around three ideas:

1. AI-driven selection. A machine-learning ranker scores each F&O stock every morning. The model is trained on years of intraday and daily price action and re-validated continuously against forward returns.

2. News correlation as alpha. Pre-market news, corporate filings, broker actions, and macro events are continuously scored and woven into the selection layer. The hypothesis: the difference between a winning and losing position is often visible in the news flow before the market opens — if you read enough sources fast enough.

3. Disciplined risk management. Position sizing, profit-locking ladders, and time-of-day exits are codified in the trading logic itself, not left to discretion.

What we're trying to achieve

Three concrete numbers are the multi-quarter target:

| Metric | Goal | |---|---| | OOS Sharpe (rolling 3-month) | Sustained > 5 | | OOS return (rolling 3-month) | Sustained double-digit | | Directional hit-rate | High-50s to low-60s % over 30-day windows |

Single-day P&L is dominated by market direction. The 30-day average is what matters.

Why this is interesting

Most retail trading tools either (a) hand you a chart and call it analytics, or (b) sell you a pre-packaged signal subscription. Neither produces a system you can actually evolve.

Project H is the third path: research-grade machine learning, news intelligence, and event-driven trading mechanics — all open and inspectable, in one repo. It's not a productised broker tool. It's the engineering log of one developer trying to answer "how good can a single person get at this?"

If the answer is "much better than expected," that's a meaningful data point about the gap between institutional infrastructure and what's actually achievable today with modern AI tooling and free data.

What this site is

A public engineering log. Not a trading-signal newsletter. No live calls, no investment advice, no paywall.

The posts here are about how the system is being built — the design choices, the AI/ML modelling decisions, the news-correlation philosophy, the experiments that worked and the ones that didn't. The exact mechanics, factor weights, and the proprietary parts of the model itself stay private. The thinking behind them is what gets shared.

What's next on the publication roadmap

• Why news correlation matters — the hypothesis that pre-market signal flow is leading-indicator alpha for intraday options, and how to capture it from free sources
• The AI/ML approach to options selection — what kind of models work for short-horizon directional ranking, what doesn't, and how to validate against forward returns honestly
• Risk management mechanics — profit ladders, time-of-day exits, and the surprising amount of P&L that lives in when you exit rather than what you bought
• Building under constraints — single-developer realities and what that forces you to design well

The 30-day live-paper-trading arc starting this week is when the engine produces its first real evidence. The blog will follow what we learn.