LangGraph
Stateful graph orchestration for LLM agents — my default for multi-step and multi-agent workflows in Python and TypeScript.
LangGraph models an agent as a state machine: typed state, nodes, conditional edges and a checkpointer that persists every step. That makes it a natural fit for Multi-Agent Systems systems, Durable Execution and Human-in-the-Loop approvals.
How I use it
- langgraph-langserve-starter — a planner → researcher → synthesizer research graph exposed as a REST API via LangServe and FastAPI, with SQLite checkpoints that resume threads over HTTP, streaming and optional Langfuse tracing.
- local-rag-chatbot — a seven-node retrieval graph (multi-query expansion, per-paraphrase retrieval, reciprocal rank fusion, grading, generate or fallback) with SQLite chat history.
- langgraph-ts-agents — a strict TypeScript port of supervisor/worker patterns with
Annotation.Rootstate, Zod-validated tools, streaming-first execution and an interrupt/resume approval gate. - Conversational AI Agents — LangGraph.js with Postgres checkpointing behind a Hono gateway in my Independent AI & Full-Stack Engineer (Freelance) work.
Credentials and courses
Multi-Agent Systems with LangGraph, part of the Agentic AI Engineering Specialization specialization — my labs above port patterns from that course.