Sentinel

01What it does

Screens options setups on a schedule, applies strategy filters, then passes every candidate through a risk layer that can only veto - never widen - risk. A confirmation server keeps a human operator in the loop, and a journal records round trips for review.

02What I built

• Strategy, execution, broker, risk, analysis, journal, and data modules with strict Pydantic models
• Scheduling with APScheduler anchored to a trading timezone
• Two-service Docker Compose deployment: an always-on confirmation web service and a scheduled screener sharing a persistent SQLite volume
• A kill switch and daily counter that survive container restarts by design
• VPS deployment tooling over SSH

03Key engineering work

• Whole-package pytest coverage gate at ~94%, with omissions documented and visible rather than silently allowlisted
• Risk sizing anchored to a cash cap; breakeven guard placement driven by journaled trade data, not guesswork
• AI-assisted analysis via the Anthropic API as an optional dependency
• Docker timezone and log-rotation configuration so missed trading windows are diagnosable