About the role
Responsibilities:
- Design and build multi-agent systems using orchestration frameworks such as LangGraph, Google ADK, CrewAI, or AutoGen, defining agent roles, tool bindings, memory, and handoff protocols.
- Develop and integrate MCP (Model Context Protocol) servers to connect agents with enterprise data sources, APIs, databases, and internal tooling.
- Implement RAG pipeline chunking strategies, embedding models, vector store selection (Weaviate, Milvus, PGVector), retrieval reranking, and context window management.
- Engineer robust prompts and system instructions; apply advanced techniques including chain-of-thought, ReAct, tool-use patterns, and structured output parsing.
- Build agent evaluation frameworks using tools like DeepEval, RAGAS, or LangFuse, designing evals for reasoning accuracy, tool-call correctness, hallucination rate, and task completion.
- Integrate with LLM providers OpenAI, Anthropic Claude, Google Gemini, AWS Bedrock, and open-source models via Ollama or vLLM, making cost-performance trade-off decisions.
- Collaborate with product and data teams to translate business workflows into agentic task graphs; iterate rapidly from prototype to production.
- Monitor agent performance in production using observability tools; implement guardrails, fallback strategies, and retry logic for reliability.
Requirements:
- 3-6 years of software engineering experience, with at least 1-2 years building LLM-powered or agentic AI applications in production.
- Strong Python proficiency FastAPI, async programming, Pydantic, and clean production-grade code practices.
- Hands-on experience with at least one agentic/orchestration framework: LangChain, LangGraph, Google ADK, CrewAI, AutoGen, or equivalent.
- Solid understanding of RAG architecture embedding models, vector databases, retrieval strategies, and context management.
- Experience with prompt engineering, structured output generation, and working with OpenAI / Anthropic / Gemini APIs.
- Familiarity with agent evaluation and observability tools like LangFuse, DeepEval, RAGAS, or similar.
- Working knowledge of cloud platforms (AWS, GCP, or Azure) for deploying AI services and comfort with Docker/Kubernetes for containerized deployments.