I build AI infrastructure that has to hold up in regulated environments.

For the last twelve months I have been building FlowDot in my spare time (https://flowdot.ai/), an agent platform where the operator stays in control: per-tool approval, immutable audit logs, hard emergency-stop, structured HITL checkpoints, on-prem deployment for sensitive workloads. Solo build across the hub (Laravel plus Node), client (React), server, mobile, VR (Unity), CLI, desktop (Electron), and an MCP server. Today I published the supervisor layer as @flowdot.ai/guardian-agent on npm under AGPL-3.0.

I was lead engineer at Obsidian SG on the VERTEX SBIR Phase 2 program. Defense training simulator deployed to AWS GovCloud, with agentic RAG on Bedrock (Claude, Llama 3 70B), real-time VR gesture recognition ML, and multi-gas chemical sensing ML running on ESP32-S3 at 10Hz with OTA hot-swaps. Customer accepted the deliverable. The Phase 2 wrapped because the company could not afford to retain me. Today is my last day at Obsidian. Everyone at Obsidian will speak highly of my performance and so will the AF customer who received the final deliverable. I am proud of the work I did at Obsidian and grateful for the opportunity to contribute to such an important project.

Twenty-plus years of full-stack work before that. Web, mobile, VR, distributed systems, embedded ML, Unity Sim2Real robotics. Led teams of three. Fifteen-plus years remote. Based in Peru, New York.