I built ENMA, a local-first agentic AI desktop assistant that delegates tasks to specialist agents and routes requests across local models. Its four-tier scoped memory gives each agent task-relevant context, while permission-gated tool execution adds audit logging and rollback protection.
For ENMA-LM, I engineered a FastAPI inference service over Ollama that streams Qwen and DeepSeek model tokens to clients via SSE. I added bounded request queues, timeouts, cancellation and GPU readiness checks to keep a single 6 GB GPU stable under concurrent load.
Through my independent software engineering projects, I’ve also built MeritOS, an AI-assisted recruitment screening platform, and prototyped INVARIANTIS, a runtime failure detection CLI/SDK. I verify my work with automated tests and develop systems across Python, FastAPI, React and Electron.

