What you’ll do

● Build the agentic orchestration layer: design and implement multi-step workflows (planning, tool use, validation, and final synthesis) using frameworks such as LangGraph or similar.

● Deliver Graph-RAG and tool integrations: connect the agent layer to our Neo4j graph, PostgreSQL state, time-series sensor data, lab results, and operational records. Improve retrieval quality, entity resolution, and query reliability.

● Make it reliable in production: implement guardrails, schema-constrained outputs, retries, fallbacks, and “needs clarification” paths. Add observability, tracing, and useful debug tooling for agent runs.

● Evaluate and improve performance: define and run evaluation suites for tool-calling accuracy, task success rate, grounding quality, latency, and cost. Reduce hallucinations with validation and tool-first strategies.

● Lead technically: write clean, well-tested code, review PRs, mentor engineers, and help set standards for the AI stack (testing, documentation, deployment practices).

Requirements (must-have)

● 5+ years of software engineering experience, strong Python, and experience shipping backend services (APIs, async workers, data services).

● Hands-on experience building LLM applications in production: tool calling, RAG, prompt design, structured outputs, and failure handling.

● Experience with an agent orchestration approach (LangGraph, LangChain, AutoGen, CrewAI, or an equivalent in-house system).

● Strong engineering fundamentals: testing, CI/CD, code review habits, observability mindset.

● Comfortable working with PostgreSQL and data pipelines; able to reason about time-series and event-style data.

Nice-to-have

● Neo4j / graph databases (schema design, Cypher, performance tuning) and Graph-RAG patterns.

● Ag-tech, IoT, sensor analytics, or decision-support systems in messy real-world settings.

● Experience with evaluation frameworks, LLMOps, and production monitoring (cost, latency, drift).

● Cloud deployment experience (Docker, Kubernetes, AWS/Azure/GCP).