Production agentic AI
for enterprise manufacturers
Enterprise RPA veteran now shipping production agentic AI — agentic pipelines, LLM decision layers, and Human-in-the-Loop approvals running in real manufacturer operations.
// current_projects
LangGraph × Qdrant × Langfuse
My working lab for agentic patterns — LangGraph orchestration, hybrid retrieval on Qdrant, and full Langfuse observability. The four years I spent shipping enterprise RPA in production are exactly what make these agents reliable: I build for the constraints of real operations, not just the happy path.
Chapter V
???
Coming next…// always building
Twilio × LangGraph × ITSM
Outbound voice AI that fires when a ITSM ticket opens. The agent calls the customer, verifies identity via 6-digit user code, collects a work note and ETA, then pushes everything back through n8n into ITSM — zero human dispatcher needed.
FastAPI × Qdrant × Redis × Langfuse
A production-grade, multi-tenant RAG API built from first principles — implementing the full 2026 RAG stack that industry leaders deploy at scale. Hybrid dense+sparse search with RRF fusion, cross-encoder reranking sidecar, async ingest queue with DLQ, streaming SSE answers, namespace-scoped RBAC, Prometheus metrics, and per-stage Langfuse tracing. 13 file formats including audio via Whisper.
// 2026 rag stack
LangGraph × Groq × FastAPI × ChromaDB × SSE
A live multi-agent system deployed on Hugging Face Spaces. A LangGraph orchestrator routes each visitor request to the right specialist: RAG for questions about my background, Email to send the CV, Calendar to book a Google Meet, Job Match to score JD fit, and Telegram for instant notifications. Every routing decision streams live to the UI via SSE.
// agents_active
Open Source
All public projects — github.com/georgelush
YouTube
Project demos & walkthroughs
Cybersecurity labs, RPA automation, Unreal Engine 5 prototypes, and C++ systems — the technical depth behind the production agentic systems I ship today.