AI Engineer for Multi-Agent Orchestration (LangGraph, CrewAI)

The Problem: We built a prototype multi-agent pipeline to automate our B2B client onboarding, but it is too brittle for production. It consists of three agents: a Research Agent (web scraping), an Analyst Agent (database compliance checks), and a Writer Agent (generating personalized onboarding plans). Right now, the system frequently breaks. We are suffering from massive data loss during state transitions between agents, infinite loops, and total system crashes whenever an LLM error occurs. What We Need: A robust, stateful orchestration layer using LangGraph or CrewAI to manage agent memory, loops, and human-in-the-loop approvals. Custom tools allowing agents to securely interact with our internal production APIs and databases. Graceful error-handling, state recovery, and token budget optimization. Requirements: You must be a strong Python engineer who understands state management and LLM memory persistence. Bonus points if you know how to securely containerize and scale these workloads in the cloud. To apply, briefly explain how you would handle state persistence when an agent fails mid-workflow. No generic templates, please.The Problem: We built a prototype multi-agent pipeline to automate our B2B client onboarding, but it is too brittle for production. It consists of three agents: a Research Agent (web scraping), an Analyst Agent (database compliance checks), and a Writer Agent (generating personalized onboarding plans). Right now, the system frequently breaks. We are suffering from massive data loss during state transitions between agents, infinite loops, and total system crashes whenever an LLM error occurs. What We Need: A robust, stateful orchestration layer using LangGraph or CrewAI to manage agent memory, loops, and human-in-the-loop approvals. Custom tools allowing agents to securely interact with our internal production APIs and databases. Graceful error-handling, state recovery, and token budget optimization. Requirements: You must be a strong Python engineer who understands state management and LLM memory persistence. Bonus points if you know how to securely containerize and scale these workloads in the cloud. To apply, briefly explain how you would handle state persistence when an agent fails mid-workflow. No generic templates, please.