What I've built.

A 3-agent pipeline (Planner, Policy, Executor) that makes autonomous decisions with real financial consequences. Multi-provider model routing across Anthropic, OpenAI, and Gemini with automatic failover. HMAC-signed receipts, a behavioral rules engine that learns from outcomes, and a self-build loop where the system proposes and validates its own new tools. Progressive autonomy model — the system earns expanded capability based on validated performance.

Policy enforcement and financial authorization layer: ActionIntent → Policy Engine → Execution Gateway → Signed Receipt pipeline. 3-plane policy engine with venue adapters for multiple trading platforms, HMAC-SHA256 receipt signing, append-only audit log, and kill switch. Designed as standalone infrastructure — any AI system that needs to handle money can plug into this.

Content scanner covering 5 injection categories, a risk scorer (0–100), and a decision guard — a fully extractable trust boundary with zero host-agent dependencies. Designed to bundle with SAMcypher as standalone trust infrastructure for any AI system that needs to make gated decisions.

AI-driven email triage and response system with voice extraction, tone calibration, calendar integration, and relationship memory. Multi-user configuration with 5 tone modes and 3 length modes. Currently in production validation.

Web scraper that identifies and flags exposed personal information across public sources. Scans, classifies, and reports sensitive data with clear boundaries on what's collected and why. Privacy audit utility demonstrating policy-aware data handling.

Multi-asset screener covering stocks and crypto with relative strength scoring, universe scanning (S&P 500, Russell 2000, Crypto Top 50), and buy signal generation with market regime awareness. Feeds live data into frAInk's autonomous trading decisions.