AI Platform Backend Performance Optimization

The platform is already in production, built on C# / ASP.NET Core micro-services with Python workers driving our OpenAI-powered features. Right now the biggest pain point is clear: performance bottlenecks inside the async processing layer are slowing down API responses and occasionally starving downstream services.

Your first mission is to trace, diagnose, and eliminate those delays. I will grant you access to the code-base, CloudWatch / Application Insights dashboards, and our Docker-based staging environment so you can reproduce the issue, attach profilers, and ship fixes behind feature flags without interrupting live traffic.

Once the backlog of bottlenecks is cleared, I would like you to:

• Harden the distributed job pipeline (message queues, retries, idempotency)
• Tune OpenAI / LLM calls for throughput and cost, caching where appropriate
• Add meaningful metrics, alerts, and dashboards so future slow-downs are caught early
• Review CI/CD workflows to be sure every change is tested, containerised, and rolled out safely

Tech you will touch every day: C#, ASP.NET Core, Python 3.x, REST APIs secured with OAuth2, Docker, AWS (ECS, SQS, RDS) – although several services also run on Azure. Data lives in PostgreSQL, SQL Server, and MongoDB. Event-driven patterns and micro-services glue it all together.

Acceptance criteria for the initial engagement
1. Background worker queues consistently drain in under one second under current peak load.
2. P95 API latency falls below 300 ms after fixes are deployed.
3. New Grafana / CloudWatch dashboards show real-time throughput, error rates, and queue depth.

If you thrive on dissecting complex systems and squeezing every last millisecond out of them, you will enjoy this challenge.