Systems I've designed, built, and shipped — from agent fleets to embedded AI hardware.
A selection of recent work across autonomous agents, trading automation, healthcare operations, and applied ML. Most are private; links go to repos, demos, or write-ups where available.
Production multi-agent orchestration platform powering Zenith — a persistent AI chief of staff coordinating a fleet of 23 specialized agents across investment, healthcare, and content workflows.
- —Built multi-agent orchestration, tool routing, shared memory, cron scheduling, and human-in-the-loop approval workflows.
- —Runs autonomously across Telegram, Discord, and Slack with specialized agents and shared context.
- —Specialized roles spanning CIO, CRO, macro, equity, crypto, and compliance.
An open-source coaching layer for Claude Code that reads local transcripts, learns developer habits, and ships ambient tips, XP, streaks, and level-ups.
- —Deterministic local-only path with zero network calls and zero token cost.
- —Ships as an npm CLI with Claude Code hooks, launchd, and cron integration.
- —Lightweight, privacy-first design that runs entirely on the developer’s machine.
A holographic AI companion running on a Looking Glass display powered by Jetson AGX Orin — embedded AI, computer vision, and holographic display hardware in one interactive system.
- —Built custom Arducam IMX708 Jetson camera driver tooling for JetPack 6.2 and open-sourced the toolkit.
- —Combined v4l2, device-tree, and Linux kernel work with real-time vision and display.
- —Novel form factor pairing conversational AI with volumetric holographic output.
A consolidated prediction-market and hedge-fund-style trading automation system — Polymarket-first scanning with Kalshi support, native agent skills, MCP tools, and approval-aware workflows.
- —Prediction scanners: sportsbook oracle, systematic NO, whale tracker, and arbitrage scanner.
- —Trading fleet with CIO/CRO-style coordination, equities monitoring, and SEC-filing watch.
- —Calibration-aware decision-making with risk controls and automated trade monitoring.
A voice AI receptionist for busy service businesses — handling inbound calls, scheduling and cancellations, structured intake, routing, and safe human transfers.
- —Designed around practical front-desk workflows, trust boundaries, and human-in-the-loop escalation.
- —Combines voice AI with browser automation for real-world booking systems.
An AI agent platform for autonomous healthcare operations — revenue cycle, clinical documentation, and practice-management workflows.
- —Designed a browser-first plus FHIR integration strategy for long-tail EHR environments.
- —Multi-agent system targeting the operational backbone of small and mid-size practices.
A real-estate price-prediction platform using an XGBoost model for price-anomaly detection.
An XGBoost model achieving 80% accuracy on NBA game predictions, served through a complete ETL pipeline and Flask API.