Shift-Left Performance Engineering for RAG and LLM Platforms
summary
Shift-left CI/CD adds an automated deviation score gate that evaluates time-to-first-token, token count, LLM call count, tool latency and error/retry rates against an adaptive 30-day baseline and blocks builds over ~2σ; implemented in eight weeks on Kubernetes with OpenAI spec–generated workloads, Prometheus/Grafana metrics, LangFuse/LangSmith/Arize tracing, PostgreSQL and Redis. Results: development-detected defects +200%, mean time to detect down ~99%, response time ~40% faster, monthly incidents down ~30–40%, feedback returns in 15–20 minutes.
AI-written summary · about
chapters (16)
- 0:00Welcome and Topic
- 0:49The Missing Gate
- 1:20Why LLM Defects Hurt
- 3:28Shift Left Framework
- 5:50Six Layer Architecture
- 8:32Deviation Score Gate
- 10:38Agentic Metrics That Matter
- 14:20Reading a Trace Signature
- 16:47CI CD Pipeline Walkthrough
- 19:16Tooling Integration Patterns
- 21:15Eight Week Implementation Plan
- 23:44Results and Productivity Gains
- 25:37Lessons and Best Practices
- 27:59Next Gen Improvements
- 30:54Demo Performance Gate
- 36:03Key Takeaways and Close
related talks
36mShifting Performance Engineering Left
23mShift-Left: A DevOps and Database Story- 23mShift-Left: A DevOps and Database Story
50mDevOps: Top Lessons Learned on Shifting Left For Accessibility and Other Experts
15mQuality Engineering Journey from Shift Left to Shift Right and Beyond
41mShift Left, Shift Right and Improve the Centre- 23mShift-Left: A DevOps and Database Story
41mPerformance Engineering in the Age of DevOps and Agile