🤖 AI Agent & Voice AI Engineer | Claude Code · LangGraph · ElevenLabs · Twilio · RAG
20+ years shipping production systems. Building voice agents, hybrid search pipelines, agentic workflows, and LLM integrations that run in the real world — not just notebooks.
- 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 hooks, MCP integrations, production guardrails.
- 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
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
- 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.
- Multi-Model AI Council (Active) — 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. - 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), 1750+ tests.
- 💼 Upwork
- 🎥 YouTube — Beyond The Developer
- 📊 Kaggle — 69 notebooks, 6 competitions, bronze medal
- 💬 HackerRank · CodeSignal
Building production AI — voice agents, hybrid search, agentic workflows, and LLM integrations. Open to senior engineering contracts and AI architecture roles.