Ultra-Low-Latency Call Metadata API using Go lang

I am building a stateless Go micro-service that will sit between Orange’s push-based feed of live call events and several AI/agent components that must query this data almost instantly for real-time analytics and automated responses. The telecom platform pushes each call record to us in JSON; from the moment it arrives I need to park it in a shared in-memory Redis instance and make it queryable through a tiny HTTP API that will be deployed on Google Cloud Run. End-to-end lookup latency must stay well under the typical network round-trip so downstream systems can react inside the ongoing call.

The service must run completely statelessly—every node simply connects to the same Redis and can be recycled at any moment without warm-up. Concurrency and connection handling therefore need to be tight and idiomatic Go. I have no interest in XML or CSV at this stage, but the design should leave room for other formats later.

Deliverables
• Production-ready Go source code with clear module structure
• Redis schema (keys, TTL strategy) plus reasoning for any sharding or stream use
• Dockerfile and Cloud Run deployment manifest; the image must start fast and respect Cloud Run memory/CPU limits
• Minimal README with build, deploy, and sample cURL commands for both ingest and lookup endpoints
• Latency test script demonstrating sub-millisecond reads within the same region

Acceptance is measured by a demo on my GCP project that shows a pushed call event being available for downstream fetch in under 10 ms at P95.