About the Role We are seeking an engineer who doesn't just "talk" to LLMs but builds autonomous, resilient systems. You will design multi-agent architecture that can reason, use tools, and recover from failures independently.   Your primary goal is to bridge the gap between "cool demos" and "production-grade reliability." You will be responsible for deploying agents across various business sectors (e.g., Finance, Operations, Customer Success), ensuring they are not only intelligent but also safe, cost-effective, and predictable in a production environment.   Key Responsibilities   ●  Agent Orchestration: Design and implement stateful, multi-turn agent workflows using frameworks like LangGraph, CrewAI, AutoGen, PydanticAI, Swarm, Haystack, or Bee Agent Framework.   ● Proactive Reliability & Guardrails: Architect systems that prevent production meltdowns. Implement circuit breakers, "human-in-the-loop" triggers, and input/output guardrails to stop infinite loops, prompt injections, and hallucinated tool calls before they reach the end user.   ● Multi-Sector Tooling: Build and maintain high-precision API integrations (tools) that allow agents to interact with diverse business systems (ERPs, CRMs, custom databases) deterministically.   ● Observability & Tracing: Set up advanced tracing (e.g., LangSmith, Langfuse, or Arize Phoenix) to debug complex reasoning chains and monitor agent trajectories in real-time.   ● Rigorous Evaluation (Evals): Develop automated "Golden Datasets" and evaluation frameworks (using RAGas, DeepEval, G-Eval, LangSmith Evaluators, or custom model-based grading) to measure agent success rates and prevent regressions before every deployment.   ● Cost & Latency Optimization: Manage the "token budget" by implementing tiered model routing (e.g., using Gemini 3 Flash for initial reasoning and Ultra for final verification) and optimizing context window usage.