Applications are closed for this role. It was originally posted 2026-07-28. It’s no longer accepting applicants — see roles PatSnap is still hiring for →, or browse the live openings below.
About this role
About the role
Patsnap's Materials team builds AI systems that help R&D scientists and engineers search, extract, and reason over materials science and patent data. You will own the agentic layer of our products end-to-end: LLM-powered agents, tools (MCPs), and the evaluation frameworks that prove they beat general-purpose AI for our customers.
You will be the AI engineer for this team — sole owner of the agentic stack, working directly with product managers, materials domain experts, and our platform team.
- Design, build, and productionize agentic systems (multi-step reasoning, tool orchestration, guardrails) for materials science search, Q&A and information extraction.
- Develop, integrate and maintain memory systems, MCP servers and agent skills in a multi-agent environment.
- Build evaluation frameworks with domain experts to measure answer quality, extraction accuracy, and retrieval performance.
- Own production reliability & observability of agents you develop.
- Advise adjacent teams on agentic and search system design; flag technical risk and feasibility during roadmap planning.
- Degree in engineering, computer science, or a quantitative/physical science — or equivalent practical experience.
- 5+ years of software/ML engineering, including 2+ years building LLM-based systems that run in production.
- You have designed evaluations for LLM/agent systems — eval sets, quality metrics, human-expert or LLM-judge pipelines — and can walk us through one (e.g., promptfoo, Braintrust, LangSmith, DeepEval, or your own harness).
- You have instrumented, monitored, and debugged live AI services (e.g., OpenTelemetry, Arize Phoenix, Langfuse, Datadog, or similar).
- Strong Python; able to independently build and deploy services.
- Search/RAG: vector databases, keyword search, knowledge graphs, reranking, hybrid retrieval
- MCP (Model Context Protocol) or agent-tool ecosystem experience
- Materials science, chemistry, or patent/IP domain exposure
- Structured information extraction from technical documents (tables, compositions, specs)
- Full ownership of a production agent stack that customers pay for
- Your evals help decide the roadmap: we build where we can measurably beat frontier general agents
- Small senior team, direct access to domain experts and real R&D users
Tech stack
LLMPhoenixPython