Enhanced Log Analytics Using Generative AI
summary
A log‑analysis pilot builds an agent network that ingests CI/CD logs, uses Grafana/Loki for retrieval and OpenAI models for classification, and exposes a Streamlit chat plus APIs to fetch job and build context. The automated classifier labels failures as infrastructure / compilation / compression and classification agent runs for every failed build, the stack includes Streamlit, Go, Loki, OpenAI farm, Kubernetes, Terraform/Crossplane, it handles ~1,000 jobs/day for a BU of ~10,000 devs, and the system processed 3 GB logs in 4 minutes while revealing cost, latency and hallucination issues (GPT‑4 timeouts, unexpected Spanish responses).
AI-written summary · about
chapters (13)
- 0:00Enhanced log analytics intro
- 1:33Autonomous agents recap
- 6:07Pilot agent for meeting minutes
- 8:58Sailing competition agent
- 12:26Agent network challenges
- 12:49Log analytics troubleshooting
- 17:05Log analytics architecture
- 18:27Model selection and costs
- 21:15Practical agent lessons
- 23:48AI solution cost comparison
- 24:54Agent frameworks comparison
- 26:43Security and scalability
- 28:31Q&A
related talks
23mAI/ML + open source observability
42mFrom Logs to Fixes: Teaching AI to Maintain Software
30mAI Agents for Incident Management
35mBeyond the chatbot, beyond the IDE, where can AI help your CI/CD
14mImproving Cloud Reliability at Scale Using Gen AI
44mAgentic Delivery, Guardrailed: AI in CI/CD & Platform Engineering (Open Source Patterns)
30mBuilding Robust, Secure LLM and Agentic AI Workflows
21mYour Deployments Are Lying. AI Knows