────────────────────────────────────────────────────────────────────
Software & Data Engineer — Python, data, and trading systems.
Tools for the moment where being confidently wrong costs more than knowing nothing.
Trading risk, French administrative rules, strategy validation — three domains where a plausible answer is the actual danger. So the tools are built to refuse. A backtest harness that kills its own edges. A skill set that will not quote a figure it cannot date. A risk dashboard that watches and never trades.
🌐 javadrazavi.fr · 💼 LinkedIn · 🛒 MQL5 Market
⭐ le-cabinet — nine roles covering French law, tax,
business finance and personal finance.
No figure is ever written from memory: 1 928 values, each carrying its official source and its
verification date, and a CI script that fails the build if a quoted number exists in no source.
Reference content is in French — the roles answer in your language.
RiskCockpit · source
Real-time prop-firm risk monitor and lot sizer. Advisor only — it reads the account and never sends an order.
DataForge · source
Exports bars to CSV and shows what is really inside them: instrument passport, per-year volume table, MQL5↔Python parity reference.
HorizonProbe · source
Is your holding time profitable by construction? The toll is fixed, the move grows as the square root of time — this measures where they cross.
ProSessionBox · source
Daily box and multi-session tracker, DST-aware, with session high/low, overlap and alerts.
quant-validation-harness
López de Prado-style validation: Deflated Sharpe, Combinatorial Purged CV, alpha/beta, bootstrap.
Built to kill false edges before capital meets them — and it killed mine, which is the point.
An edge that survives only the test that discovered it is not an edge, it is a memory of the data. The same holds for a rate quoted from memory, or a dashboard that reassures instead of measuring. Every repository here is an attempt to make a tool say "I don't know" out loud, on time.
🐍 Python · MQL5 · TypeScript / React · Rust (Tauri) · SQL · ELK · Docker · CI/CD
📬 Building something in Python or data? I am listening — collaboration, tooling, or a system that has to be right before it is fast. Tell me about it on LinkedIn.