From Noisy Alerts to Smart Fixes
summary
Company alert noise turns into alert-to-PR codegen that parses dbt lineage, generates SQL fixes, runs dbt recompiles on Snowflake, and opens a GitHub PR for human review. It handles nullness and fan‑out cases across thousands of models for a ~30‑person data org, but requires careful Snowflake perms, tooling to run queries, and explicit success metrics — namely dbt unit test validation. A separate prompt-refiner feedback loop uses review labels to iteratively improve prompts; outcome: useful for bounded ETL debugging, not for broad coordination.
AI-written summary · about
chapters (8)
- 0:00Introduction and speaker intro
- 1:27Framework for AI task prioritization
- 5:01What is an analytics engineer
- 8:13On-call workflow for analytics engineers
- 10:24Building the alert-fixing agent
- 13:12Code generation success criteria
- 16:24Feedback loops and prompt refinement
- 17:28Q&A session
related talks
30mPrompt to Performance Speeding Up Database Administration with AI and Claude Code
29mOne-Click Code Fix: Securing Code Using AI
31mAlerting for Distributed Systems—A Tale of Symptoms and Causes, Signals and Noise
26mLet Them Eat Bugs: Practical Showcase of Agentic Issue Resolution
41mImproving DevOps workflows with GenAI
32mOpenTelemetry, databases, and machine learning
56mWhy Auto-Fixing Vulnerable Code Needs More Than Good Intentions
23mFrom Blind Spots to Merged PRs: Continuous Agentic Performance Optimization