I ship production web apps and GenAI systems end to end. 5+ years across JavaScript/TypeScript, Python, and Kotlin, with Go, Ruby, and Rust in the mix. Right now I split my time between frontier-model evaluation work (DeepMind Gemini via Turing, Meridial via Invisible, Revelo) and architecting AI products for clients.
If you've ever wondered whether a voice agent can actually do the thing instead of just talking about it — I'm probably your kind of engineer.
Four portfolio projects, scoped to be production-grade rather than demos. One is live, three are mid-build.
|
Production-grade voice agent over hybrid RAG with idempotent action handling, YAML-driven domain config, eval harness in CI, full OTel tracing. WebRTC · Pipecat · Qdrant · FastAPI · Langfuse |
⚙️ workflow-engine (in progress)Durable workflow engine in Go with a Remix-based time-travel debugger for inspecting in-flight runs. Go · Remix · PostgreSQL · OTel |
🎨 collab-canvas (in progress)CRDT-backed real-time collaborative whiteboard with offline-first sync. TypeScript · Yjs · WebSocket · IndexedDB |
📊 eval-lab (in progress)Open LLM evaluation platform with statistically grounded regression detection. FastAPI · DuckDB · pytest · React |
- Turing — POD Lead (Jun 2024 → present). DeepMind Gemini post-training. RLHF failure-pattern analysis, golden-solution authoring, a multi-month cornerstone code-gen benchmark with Dockerized behavioral verification, and Agent Skills tasks for Gemini 3.1 Pro calibrated to 4-of-5 fail rates across 7 domains. Mentored 12+ contributors.
- Invisible Technologies (Meridial AI) — Senior AI SWE & QA Lead (Oct 2025 → present). SWE-bench-style Kotlin tasks on pinned OSS commits, verifier-grade coding problems across 8+ languages, pairwise agent-trajectory eval for DeepMind.
- Crossover — AI First SWE (Aug 2025 → Feb 2026). Cut Khoros platform EC2 spend 50% by migrating to AWS Graviton ARM64 via Ansible playbooks reading golden AMIs — runtime-config re-architecture, not container rebuilds.
- Mutekra — Full Stack Engineer (Mar 2025 → Feb 2026). Shipped Musymptra (clinical symptom-triage with OCR + LangGraph-gated questionnaires), Musupportra (ISP Tier 1 orchestrator across voice/chat/WhatsApp), Muswarm (real-time multi-agent visualization with MCP sub-agent orchestration).
- Revelo — Frontend AI Evaluator (Mar 2026 → present). Figma → vanilla HTML/CSS/JS golden-solution components for frontier LLM training data.
AI & Agents
Languages
Frontend
Backend
Data
Cloud / DevOps
Open to senior AI / full-stack roles · Remote-friendly · Let's talk