Chat with Me and Get the PromQL You Want
Talkat KubeCon + CloudNativeCon + Open Source Summit China 2023 · on Wednesday, 2023-09-27· at 13:55· 3层 305B会议室| 3F Room 305B·
summary
A chatbot converts natural language into PromQL using LLM APIs plus a local knowledge base and vector search, with offline Q&A examples turned into embeddings and online top-K retrieval before prompting. Early experiments show <5% accuracy from docs, ~20% from QA+AI explanations, and chain-of-thought prompts raising correctness to ~70% and then to about 76.6% across 14 scenarios. System plugs into ARMS/Grafana, is extensible by adding ~5 examples per new scenario, and returns PromQL plus reasoning steps.
AI-written summary · about
chapters (11)
- 0:00Natural language to PromQL
- 1:15Why natural language matters
- 4:28Building a PromQL tool
- 9:56PromQL generation experiments
- 12:42Chain-of-thought prompting
- 17:03PromQL chatbot architecture
- 18:51Live demo: PromQL chatbot
- 21:56Current accuracy and capabilities
- 24:05Natural language to SQL in observability
- 26:15Vision for automated chatbot
- 26:37Key takeaways
related talks
37mAI in SRE: Unlocking Prometheus Insights with Natural Language
46mPractical End-to-End AI Development using Prompty and AI Studio
29mInside a PromQL Query: Understanding the Mechanics
27mEnhancing AI-Driven Quality Assurance
59mMaking the most of artificial and human intelligence for data science
31mMachine Learning-Enabled Chatbots
19mDo We Still Need to “Observe”? the Future of AI & Observability
39mPragmatic AI: Reshaping the Product Opportunity Tree