About the role
Responsibilities:
- Design and build multi-step agents using LLMs, tool use, retrieval, and memory in frameworks like LangChain, LangGraph, or similar.
- Architect single-agent and multi-agent systems: subagents, routing and handoffs, planning, and state management for long-running tasks.
- Build evaluation frameworks for agent behavior, LLM-as-judge, and deterministic evaluators and use them to iterate on prompts and architecture.
- Integrate agents with client systems and data sources: APIs, databases, vector stores, and MCP or tool integrations.
- Deploy agents to production and set up observability, tracing, logging, and error analysis to monitor real-world behavior.
- Work directly with clients and domain experts to translate a workflow into an agent spec and iterate on it based on production feedback.
Requirements:
- 3-7 years of experience in backend or full-stack engineering, with a track record of shipping production systems.
- Hands-on experience building agents with LLM frameworks (LangChain, LangGraph, or similar), including tool use and multi-step reasoning.
- Experience with prompt engineering and evaluation frameworks and iterating on agent behavior using real feedback.
- Strong proficiency in Python and/or TypeScript, APIs, and system design; cloud experience preferred.
- Comfort taking ownership of a project end to end, from architecture to a client using it in production.
- High ownership, strong execution, and the instinct to improve systems without waiting to be asked.
Nice to Have:
- Experience with RAG patterns, vector stores, and knowledge retrieval.
- Experience with agent observability tooling (LangSmith or similar) and debugging agent behavior in production.
- Experience working directly with enterprise clients on technical scoping and delivery.
- Exposure to RLHF, SFT, or post-training workflows.