AI-augmented systems builder. I design systems and build them end to end, using AI as my execution layer so one person delivers what usually takes a small team. I own the fundamentals (Python, systems design, debugging); AI speeds up delivery, it does not replace the thinking.

Give me a fuzzy goal and I turn it into a real architecture and working, deployed software with tests, monitoring, and safe-rollout machinery. I ramp on a new domain or API in days, work fully remote, async, and text-first.

• Automation & backends: bots, scrapers, data pipelines, AI agents, lightweight HTTP services, dashboards, SQLite-backed state, systemd deployment.
• Trading & data systems: probabilistic modeling, calibration against real outcomes, backtesting, drift monitoring, champion/challenger rollouts.
• Web3 security research: attacker-style review of smart contracts, access-control and trust-architecture flaws, proof-of-concept validation. 17 valid findings on CodeHawks First Flights, each with an exploit that runs. Human-directed, AI-augmented: I set the hypotheses, read the code, and validate every finding.

scout-osint A single-operator OSINT and research aggregation agent. Give it a domain, company, username, email, or topic and it collects public information through 17 collectors, correlates identifiers into a confidence-scored entity graph, has an LLM write a structured report, and serves it on a local dashboard. About 16,300 lines of Python across 115 commits, with 733 tests and all network mocked. Standard-library first, SQLite encrypted at rest, prompt-injection isolation in the synthesis layer, and a written spec with hard ethics constraints: public data only.

codehawks-findings Selected smart-contract security findings from CodeHawks First Flights, published as full write-ups with runnable Foundry and Anchor proof-of-concept exploits. 17 valid findings overall (8 High, 7 Medium, 2 Low) across three contests; the six published here were picked to span distinct vulnerability classes rather than to pad a count. Profile: profiles.cyfrin.io/u/augustwh11.

weather-prediction-market-bot An automated Polymarket trading system, about 32,000 lines of Python built solo across 54 commits. Forecasts station-level temperature from multiple meteorological models, builds calibrated probability distributions over market bins, and enters positions with positive expected value. Engineered for safe operation: champion/challenger, shadow runtime, dual-write, gates, drift monitoring, systemd deployment, rollback runbooks, and a 1,100+ line test suite. Runs in paper mode by design, as an engineering and research project.

postmortem-key-compromise-collateral-trust A security post-mortem of an admin-key compromise on the Echo protocol (Monad): unauthorized mint and chained exploitation through lending protocols, with a trust-architecture checklist (single EOA mint authority, missing timelock, and more).

technique-eth-call-state-override A write-up on using the eth_call state override to binary-search threshold values inside contracts without sending any on-chain transactions.

17 valid findings on CodeHawks First Flights so far; next up is live competitive audits, where findings are ranked against production protocols. Also taking on remote automation, backend, and OSINT/research-tooling work. Advanced hands-on crypto user since 2017.