🤖 AI Architect & Voice AI Engineer | Production AI · Governance · Claude Code · LangGraph · ElevenLabs · Twilio · RAG
20+ years shipping production systems. I build AI that goes to production — with evaluation pipelines, human-in-the-loop oversight, and governance built in from day one. Not demo notebooks. Governed, tested, production-ready systems.
- AI governance & human oversight — Every agent I ship has bounded authority, HITL escalation paths, and audit-ready logging. Not a checkbox: built into the architecture from day one. Demonstrated across the Claude Code, n8n, and LangGraph series.
- Evaluation before deployment — 1,750+ automated test scenarios validated on a production voice ordering system; pytest coverage across all agentic pipelines; structured output validation with Pydantic. AI that is tested, not trusted blindly.
- Secure agentic systems — MCP server development with tool-permission scoping, prompt injection awareness, and production guardrails. See: Claude Code in Production series.
- Voice AI Systems — Sub-200ms ElevenLabs TTS over bidirectional WebSocket, Twilio Media Streams & ConversationRelay telephony bridges, LangGraph barge-in with Redis checkpointing.
- Production RAG & Hybrid Search — BM25+dense hybrid retrieval (Qdrant), RRF fusion, cross-encoder reranking, FastAPI serving. pgvector, Pinecone, ChromaDB.
- AI Agents & Orchestration — LangGraph multi-agent pipelines, PydanticAI, multi-model councils (Claude + Gemini + GPT-4o in parallel async), MCP server development.
- Claude Code & Agentic Dev Workflows — CLAUDE.md architecture, PostToolUse/PreToolUse/Stop hooks, MCP server development, production guardrails. 4-episode published series.
- n8n AI Agent Engineering — Self-hosted n8n agents with persistent Redis memory, RAG pipelines, HITL email gates, Slack bots, multi-agent orchestration. 4-episode published series.
- LLM Integrations — Anthropic Claude, OpenAI, NLP-to-SQL (Vanna.ai), structured output pipelines, prompt caching, token optimization.
- Legacy Modernization — .NET Framework → .NET 10, Silverlight → OpenSilver/WebAssembly, monolith → FastAPI microservices (Strangler Fig, Circuit Breaker, Saga, Outbox).
- Backend & Full-Stack — Python/FastAPI, .NET/C#, Node.js/TypeScript, PostgreSQL, Docker, AWS (ECR, EC2, Cognito, S3).
Production voice AI series — ElevenLabs bidirectional WebSocket TTS (sub-200ms), Twilio Media Streams vs ConversationRelay telephony bridge (Mu-law/AEC), multi-modal Vision + Voice design, LangGraph Supervisor with Redis barge-in + speculative TTS.
Python FastAPI ElevenLabs Twilio LangGraph Redis OpenAI Vision
▶ Ep1 · Ep2 · Ep3 · Ep4
Production RAG series — BM25+dense hybrid retrieval, RRF fusion, FastAPI serving layer, cross-encoder reranking. Benchmarks and architectural tradeoffs across all stages.
Python Qdrant FastAPI BM25 RRF Cross-encoder reranking
▶ Ep1 · Ep2 · Ep3
Claude Code in production — CLAUDE.md architecture, PostToolUse/PreToolUse/Stop hooks, MCP server development with FastMCP, context management patterns for long sessions.
Claude Code Python MCP Bash hooks
▶ Ep1 · Ep4
n8n AI Agent Engineering series — self-hosted n8n agents with persistent Redis memory, RAG pipeline with Qdrant, HITL email gate, Slack bot, multi-agent orchestration. Production self-hosting + supervisor patterns.
n8n Python Redis Qdrant Docker
▶ Ep1 · Ep2 · Ep3 · Ep4
AI Engineering for Real Apps series — LangGraph multi-agent report pipeline with structured Pydantic outputs, Anthropic Claude, pytest coverage.
Python LangGraph Anthropic Claude Pydantic pytest
All 7 microservices migration patterns (Strangler Fig, API Gateway, Sidecar, Outbox, Circuit Breaker, Event-Driven decoupling) built with Claude Code agentic workflows.
Python FastAPI Kafka SQLAlchemy Docker Claude Code
Practical .NET 10 legacy migration lab — RAG integration, MCP server wiring, clean architecture without pretending the legacy system disappears overnight.
C# .NET 10 RAG MCP
- ai-decision-room (public — contract closed) — Claude Opus (Chairman) + Gemini Flash (Analyst) + GPT-4o (Risk Officer) running parallel async Python via
asyncio.gather, synthesized by Claude Sonnet. Vue 3 SPA, Docker + Nginx, deployed to Mac Mini M4 via Tailscale within a 2-week sprint. - Media Distribution Platform (AI Reporting) — NLP-to-SQL with Vanna.ai, ChromaDB RAG chatbot, FastAPI + Anthropic Claude. Taken over mid-development; shipped LLM features in production.
- Voice Ordering System Architecture Review (NDA) — FSM+LLM hybrid, LiveKit Agents (Deepgram Nova-3 + GPT-4o-mini + Cartesia Sonic-2), OpenAI Realtime API fallback analysis ($0.82/call vs $0.08/call), 1,750+ tests.
- 💼 Upwork
- 🎥 YouTube — Beyond The Developer
- 📊 Kaggle — 69 notebooks, 6 competitions, bronze medal
- 💬 HackerRank · CodeSignal
Building AI that ships to production — governed, evaluated, human-in-the-loop. Voice agents, hybrid search, agentic workflows, MCP servers. Open to senior engineering contracts and AI architecture roles.