About the role
Key Responsibilities· Design and implement agentic systems: single-agent and multi-agent architectures (planner/executor, supervisor-worker, routing, reflection, critique, task decomposition).· Build robust tool-using agents: function calling, tool schemas, tool authorization, retries, rate limiting, and sandboxing.· Implement RAG + memory patterns: retrieval strategies, hybrid search, context assembly, long-term memory, and grounding/citation behaviors.· Develop workflow orchestration for agent execution (state machines/graphs), concurrency controls, and deterministic execution where possible.· Productionize GenAI services: APIs, background jobs, streaming responses, caching, and cost/latency optimization.· Establish agent evaluation: golden sets, simulation-based evals, LLM-as-judge with mitigations, task success metrics, regression testing.· Build observability and safety: tracing, token/tool telemetry, anomaly detection, prompt injection defenses, data leakage prevention, policy enforcement.· Collaborate with product, security, and platform teams to deliver enterprise-ready solutions and integrate with internal systems (data, identity, workflow).· Mentor engineers, set coding standards, and contribute to architecture reviews and technical roadmaps.Required Qualifications· 6+ years software engineering experience; 2+ years building LLM/GenAI systems in production.· Strong programming skills in Python (required) and/or TypeScript/Node.js.· Hands-on experience building agents (tool calling, planning, routing, multi-step workflows) beyond simple chatbots.· Solid understanding of prompting, context window management, grounding, hallucination failure modes, and mitigation strategies.· Experience with RAG: embeddings, vector databases, chunking strategies, hybrid retrieval, re-ranking. Proven ability to ship production services: Docker/Kubernetes, REST/gRPC, CI/CD, monitoring, incident response basics.· Strong data/security instincts: secrets management, PII handling, secure tool access, least privilege