~/projects
Selected Work
Projects that represent how I think about AI systems, product design, and engineering leadership. Each one taught me something worth showing.
Multi-Agent OpenClaw Orchestration
Built a role-based multi-agent system with explicit orchestration, specialist routing, durable memory, and handoff workflows. The goal was not impressive demos. It was reliable AI work across sessions.
Jarvis: Hermes AI Operating System
A persistent, always-on AI operating system built on Hermes Agent. Layered memory, a 125-skill library, a 590-file Obsidian knowledge vault backed by GitHub, and cron-driven autonomous workflows. Accessible from anywhere via Telegram.
Employee Communication Simulator
Built a simulation platform for testing how leadership messages land across a 450-person org. Weighted personas, Monte Carlo scenario analysis, multi-provider LLM integration, and executive-ready reporting in PDF and PowerPoint.
Browser Workflow QA Agent
An AI agent that performs structured QA on live websites. Navigate pages, audit console state, capture desktop and mobile screenshots, produce evidence-backed PASS/WARN/FAIL reports. Triggered by a single Telegram message.
Agent Trust Control Plane
Built a local governance layer for AI agents. Every proposed action is classified as allow, approval_required, clarification_required, or blocked. YAML policies, deterministic risk classification, human approval gates, and JSONL audit trails.
OpsCore Dashboard
A real-time operational dashboard for monitoring containerized AI services on a VPS. Python metrics agent on the server, Next.js frontend on Vercel, remote visibility into CPU, memory, disk, and container health.
Durable Memory for AI Agents
A reliability engineering case study on making long-running AI systems trustworthy. Explicit source-of-truth artifacts, retrieval rules, and handoff workflows. The kind of work that matters on Tuesday morning, not just when everything is going well.