- Multi-Agent Orchestration : Design and implement complex, stateful multi-agent workflows using LangGraph will manage cycles, human-in-the-loop triggers, and persistent state.

- Agentic Tooling : Build robust tool-calling interfaces that allow agents to interact.

- Advanced RAG : Move beyond basic vector search. Implement "Agentic RAG" where agents can iteratively search, critique, and synthesise information from massive psychographic datasets.

- Evaluation and Guardrails : Build the "Eval" pipeline. Use LangSmith or custom frameworks to measure agent trajectory quality and prevent hallucinations.

- Production Deployment : Take at least one core agent from concept to a scaled, production-grade service that handles high-concurrency requests.

Requirements :

- 1-3 years of experience with at least 1 year dedicated to building and scaling LLM-powered applications in production.

- Python Mastery : You write clean, asynchronous Python code and are a pro with FastAPI/Pydantic.

- Deep LLM Understanding : You understand context windows, tokenisation, and the nuances between models (GPT-4o vs. Claude 3.5 vs. Llama 3).

- Framework Expertise : High proficiency in LangChain and LangGraph, understanding how to manage "State" and "Nodes" in a cyclic graph.

- Tooling : Experience with n8n for rapid workflow automation and Vector DBs (Pinecone, Weviate, or Qdrant).

- The "Tinkerer" DNA : A portfolio (GitHub/Blog) that shows you've experimented with autonomous agents, autonomous research, or custom cognitive architectures.