What I build (deliverables)
Exchange integration
REST + WebSocket, multi‑venue adapters, normalized market data, order & position management.
Backtesting infrastructure
Train/validation/holdout discipline, realistic fees + spread + slippage, repeatable experiment runs.
Execution systems
Order lifecycle tooling, reduce‑only safety, staging + replay verification, failure‑mode hardening.
Microstructure & fill quality tooling
L2 analysis, fill toxicity measurement, adverse selection diagnostics, latency/queue effects.
Real‑time data pipelines
WebSocket ingestion → durable logs, systemd persistence, replay, multi‑asset logging at scale.
How I work
Architecture audit (fixed scope)
3–5 days. Review data, backtests, execution, and ops. Deliver a prioritized roadmap + acceptance criteria.
Build sprint (dedicated)
2–6 weeks. Ship one production subsystem end‑to‑end (integration / pipeline / execution / tooling).
Ongoing retainer
Embedded infrastructure support: ship improvements, keep uptime high, respond to edge cases.
Recent work
Case study: Adverse selection measurement
- Problem
- A perp market‑making system looked positive in simulation, but real‑world fill quality was unknown.
- Built
- Real‑time L2 analysis pipeline capturing queue depth, touch‑through rates, and fill toxicity across 229 perpetual markets. Durable logging via systemd with auto‑restart + replay.
- Result
- Quantified that ~78% of simulated fills would be adversely selected live — preventing deployment into a structurally losing strategy.
Case study: Automated research pipeline
- Problem
- Manual strategy evaluation was slow and overfit‑prone; no systematic way to test large grids with proper methodology.
- Built
- Automated pipeline: ingestion, strategy generation, train/validation/holdout splitting, realistic cost modeling, and structured reporting.
- Result
- Evaluated 600+ variants across 3 asset classes in < 24 hours with reproducible runs and no leakage between splits.
Case study: Real‑time flow monitoring
- Problem
- Detect and analyze forced liquidation flow in real time across a major perp exchange.
- Built
- Multi‑service monitoring tracking vault activity across 7 subaccounts and 229 markets, with cascade detection and Telegram alerting. Deployed as persistent systemd units.
- Result
- Operational within hours; captured high-frequency telemetry with high uptime during the observation window.