Job description
Design and build multi-agent and agentic workflows using Lang. Graph or comparable orchestration frameworks — state machines, tool use, memory, retries, and human approval gates. Develop production backend services in Python and/or Java, with clean API design, sound data modelling, and real test coverage. Build front-end interfaces (React or similar) that expose agent behaviour clearly — traces, intermediate steps, confidence, and points of human intervention.
Implement retrieval pipelines: chunking, embeddings, vector stores, hybrid search, and evaluation of what actually improves answer quality. Instrument everything. Evaluation harnesses, tracing, cost and latency monitoring, regression testing of prompts and graphs. Deploy and operate on cloud infrastructure with containers and CI/CD.
Work directly with architects and clients — translating requirements into working systems, and pushing back when the requirement is wrong.
Requirements
4+ years building production software in Python and/or Java. Strength in one and working fluency in the other is fine. Demonstrable experience with agentic frameworks — Lang. Graph, Lang. Chain, CrewAI, Auto. Gen, the OpenAI or Anthropic SDKs, MCP — including systems that made it past the prototype stage. Solid full stack range: REST and event-driven APIs, relational and vector databases, a modern JavaScript framework.
Understanding of the failure modes: hallucination, prompt injection, tool misuse, runaway cost, non-determinism — and the engineering patterns that contain them. Comfort with Docker, Git, CI/CD, and at least one major cloud provider. Clear communication in English, written and spoken.
Nice to have
Financial services or other regulated-domain experience. Model evaluation and observability tooling (Lang. Smith, Langfuse, Weights & Biases, or equivalent). Kubernetes, infrastructure as code, or platform engineering background.