About the role
THE ROLE
Architecture & System Design
- Design multi-agent architectures (e.g., Supervisor-Worker, Hierarchical Teams) capable of breaking down complex user queries into sub-tasks.
- Define the state management strategy to ensure agents retain context, memory, and user intent across long-running workflows.
- Architect robust Retrieval-Augmented Generation (RAG) pipelines that allow agents to query proprietary data with high precision.
- Select and integrate appropriate LLM orchestration frameworks (e.g., LangGraph, AutoGen, CrewAI) based on use-case requirements.
Engineering & Development
- Implement tool-use capabilities (function calling), enabling agents to interact with internal APIs, databases, and third-party SaaS platforms safely.
- Develop guardrails and steering mechanisms (e.g., NeMo Guardrails, LMQL) to ensure agents stay "on-rails" and avoid hallucinations or unsafe actions.
- Optimize prompt engineering strategies (Chain-of-Thought, ReAct, Tree of Thoughts) for maximum reliability and minimum latency.
- Oversee the transition from prototype to production, ensuring code is modular, testable, and scalable.
Production Operations (LLMOps)
- RayImplement evaluation frameworks (e.g., Ragas, TruLens, DeepEval) to quantitatively measure agent performance, accuracy, and hallucination rates before deployment.
- Design observability dashboards (using tools like LangSmith, Arize, or Datadog) to trace agent reasoning steps, token usage, and latency in real-time.
- Manage cost and performance trade-offs, implementing caching strategies and selecting the right model mix (e.g., routing simpler tasks to smaller/cheaper models like GPT-4o-mini or Llama 3).
