AI Architect required for Multi-Agent Workflow, Tenant Isolation & Cost Optimization

We run a high-volume commercial services and logistics business handling thousands of weekly requests across hundreds of client accounts and vendor networks. Our operations desk is bottlenecked by unstructured emails, quote requests, and status updates. We have an early prototype for an automated multi-agent workflow, but we need an experienced architect to solve three critical production hurdles: Token & Cost Spikes: Unbounded prompt loops and redundant LLM calls are driving up inference bills. Tenant Isolation: We manage competing client accounts and cannot risk rate sheets or proprietary data leaking across accounts in shared memory/vector stores. Agent Governance: Autonomous agents calling external internal APIs occasionally loop, fail schema validation, or execute actions without an audit trail. We need an Enterprise AI Systems Architect for a focused 8-hour paid architecture audit & scoping sprint. A successful sprint will lead directly to a fixed-price contract to build out the full system. Deliverables for the 8-Hour Sprint: Multi-Agent Routing & Guardrails: Map a deterministic supervisor-agent routing pattern (LangGraph / custom state machine) with loop limits and human-in-the-loop gates for external API calls. Tenant Segregation Blueprint: Define strict tenant data isolation in vector search (pgvector/Qdrant) and session memory so client data is cryptographically or namespace-partitioned. Cost Optimization & Caching Strategy: Specify model tiering (fast/low-cost models for routine parsing, frontier models for complex reasoning) and semantic caching to slash token spend. Phased Implementation Roadmap: Deliver a concise 1-page architecture schematic with fixed-bid milestones for our engineering team to execute. Stack Requirements: Custom Python & API-first backends (FastAPI, LangGraph/LlamaIndex, PostgreSQL/pgvector). Zapier/Make experience required - stateful execution, deterministic schema validation, and full code ownership. To Apply, Answer Directly: How do you technically enforce tenant isolation in vector search to prevent cross-account leaks? What guardrail do you use to stop autonomous multi-agent loops from runaway API and token costs? State your fixed price or hourly rate for this 8-hour sprint.