Build a Go LLM gateway in front of our Python RAG service (6 weeks, likely extension)

We're a SaaS running Go microservices on Kubernetes (EKS), PostgreSQL and Kafka. This spring we shipped an AI assistant built on Python/FastAPI and OpenAI. It works but it's fragile under load, we have no rate limiting per customer, and our OpenAI bill is out of control. We need a proper gateway in front of it. Scope: - Go service that sits between our app and the LLM providers: routing across OpenAI, Anthropic and Bedrock, streaming, retries and fallback, per-customer token budgets, response caching - Hardening the existing FastAPI RAG workers: job queue, timeouts, structured logging. Not a rewrite - Prompt and retrieval regression checks in our GitHub Actions pipeline so changes get tested before deploy - Helm chart, Docker image, deployed to our EKS cluster, Grafana dashboard for latency and cost - Handover doc for our two backend engineers Must have: 5+ years Go in production, Docker and Kubernetes beyond local, PostgreSQL, one LLM feature shipped to real users, enough Python/FastAPI to work in the existing service. Nice to have: built an LLM proxy before or used LiteLLM/Portkey and know their limits, pgvector, Kafka. After this there's a longer engagement on the table