Jimmy M.

🧭 **I design and build agentic systems that survive production — observable, evaluable, and maintainable by your team after I leave.** 🏅 **Expert-Vetted | 100 % Job Success | Physics PhD, $90K+ earned. Recent: VIVIA (real-estate multi-agent platform, WhatsApp + voice + CRM) and a Neo4j-backed RAG pipeline rated ★5.0 Agents fail where teams don't look: prompt fragility, silent regressions, untraced failures, no fallback paths. That's the first thing I plan for. 🏗️ **Architectural scope I cover** - Multi-agent orchestration with LangGraph — explicit state machines, checkpointing, human-in-the-loop - GraphRAG on Neo4j — entity resolution, hybrid retrieval, explainable reasoning paths - Agent evals & observability — LangSmith / custom tracing, regression suites, cost dashboards - Production deploys — vLLM / Docker / Kubernetes, load-balanced inference, fallbacks across providers - Integration layer — WhatsApp, Twilio voice, CRM APIs, MCP servers, webhook resilience 🔬 **Background** PhD in Physics + years as a quantitative researcher. I treat agent behavior as a measurable system — I run experiments, log everything, and challenge claims that don't come with numbers. ✅ **Recent work** - **VIVIA** — Multi-agent real-estate platform: channel-agnostic core, per-agency scoping, lead follow-up sync, voice calls, CRM integration. Led modular refactor from monolith to clean separation of concerns. - **LLM/VLM fine-tuning & deployment** — custom load-balanced GPU platform, $10K+, ★5.0. - **AI Compliance system** — LLM guardrails, evaluation harness, and audit trail for regulated-industry workflows (i.e., full traceability). - **Neo4j calendar-event ingester** — production graph pipeline, ★5.0. 🛠️ **Stack** LangGraph · LangChain · LlamaIndex · Neo4j · Qdrant/Pinecone · Supabase · vLLM · FastAPI · Python/Cython/C++ · Docker · Twilio · MCP · WhatsApp/Notion SDKs If you're scoping an agent or RAG system and want someone who'll push back on the architecture before writing code — **let's talk.** 📬