I build the full quantitative-finance stack — a matching engine, the pricing math, a backtester, and a language to write strategies in — then use it to test trading ideas honestly in a monthly research lab, publishing what the data really says (null results included).
| 🧪 Tests passing | 🔬 Research papers | 🏦 Quant repos | 🧮 Pricing engines | ⚡ Order book |
|---|---|---|---|---|
| 140+ | 3 · reproducible | 6 | 3 cross-validating | 165k orders/sec |
🔬 Latest from Martingale: Note 004 — How effective are liquidity-grab / FVG setups, statistically? → Across 2,903 days of SPY, no edge: at 1 day nothing is significant; the 5-day "significant" results are just market drift, and the bearish patterns are followed by the largest positive moves — the opposite of the claim. read the paper →
🧫 Currently researching: Note 005 — Where do SPY's returns actually come from — overnight vs. intraday?
A question, an experiment, an honest answer — including the null ones. Most projects claim to find edges; this lab rigorously tests whether claimed edges are real, and reports what the data actually says. Every note is a hypothesis fixed in advance, a reproducible experiment with no lookahead, and a written paper — findings that stand on their own, especially when the answer is "it doesn't work."
| # | Question | Honest finding |
|---|---|---|
| 001 | Do backtests overstate performance? | A 1-line lookahead bug inflates a strategy's Sharpe by +1.12; cherry-picking 337 strategies on random data fakes a 0.77 Sharpe that flips to −0.54 out-of-sample |
| 002 | Does volatility predict next-day direction? | Across 2,881 days of SPY, today's volatility → tomorrow's return correlation is +0.035 — no meaningful directional edge (R² ≈ 0.1%) |
| 003 | Momentum vs. mean reversion, after costs? | Over 4,916 days, neither beats buy-and-hold — but each is a regime bet: mean reversion earns a 1.07 Sharpe in bear markets, then its ~6× turnover lets 5bp costs turn it negative |
📄 Read the papers · 🔁 Reproduce any note · 📌 Cite this work · open to contributors
🤖 Auto-updated daily by a GitHub Action — a backtest of my own quantsim engine, refreshed every morning. (paper research, not investment advice)
| as of | strategy | buy & hold | verdict |
|---|---|---|---|
2026-09-23 on sample series (offline) |
+60.9% · Sharpe 0.72 · maxDD -21.3% | +14.6% · Sharpe 0.22 | ✅ beating buy & hold |
Two ideas from my repos, brought to life as pure self-animating SVG (SMIL) — regenerated daily by the same Action. Left: a limit order book matching engine. Right: Monte Carlo option pricing converging to Black–Scholes.
I got curious about how markets actually work — so I built the whole stack to find out: a matching engine, the pricing math, a backtester, a live trading bot, and finally a language to write strategies in. Then I started a research lab to point that stack at real data and test popular trading ideas the honest way — hypothesis first, no lookahead, realistic costs, and the null results published alongside the wins. I like zero-dependency code, tests that assert real properties (not just "it runs"), and work you can reproduce in under a minute.
- 💸 Quantitative finance — backtesting, options pricing, market microstructure, honest strategy evaluation
- 🔬 Research — reproducible experiments, dual-licensed papers, methodology over hype
- 🤖 AI developer tooling — MCP servers and infrastructure for LLMs
| Project | What it does | Built with |
|---|---|---|
| 🔬 quant-research | Martingale — monthly, peer-reviewable quant research. Reproducible experiments + written papers that report the honest answer, null results included. Open to contributors | Python |
| 📈 quantsim | Full quant stack — backtesting engine, price-time-priority order book with market-impact execution, Monte Carlo risk analytics, and a live paper-trading bot that commits its P&L to git | Python · NumPy |
| 🏛️ exchange-simulator | Agent-based market where fat tails, volatility clustering & flash crashes emerge from autonomous traders — statistically verified | Python |
| 📜 quantlang | A programming language for trading strategies — hand-written lexer, parser & interpreter; compiled output proven bitwise-identical to hand-written Python | Python |
| 🧮 optionslab | Options pricing with three independent engines that cross-validate to 4 decimals — Black–Scholes, binomial trees, Monte Carlo — plus Greeks & implied vol | Python |
| 🧭 pathfinding-visualizer | Watch A*, Dijkstra, BFS & Greedy race across a grid you draw — 60fps canvas, zero deps | JavaScript |
