Experiment
AI DBA Assistant
An experimental AI-powered assistant that reads SQL Server health data and suggests troubleshooting directions, built to explore how AI-assisted engineering fits into day-to-day DBA work.
Source code not published
Problem Statement
Interpreting wait statistics, blocking chains, and DMV output still requires DBA experience - but a lot of the first-pass triage (what does this wait type usually mean, what should I check next) follows recognizable patterns. This is an experimental project exploring whether an AI assistant, given the same health check output a human DBA would look at, can offer a useful starting point for troubleshooting rather than a full-blown production tool.
Architecture
Technologies Used
- Python for collecting and formatting SQL Server health data.
- Prompt engineering to turn DMV/wait-stat output into useful context for an AI model.
- Claude models and GitHub Copilot used both to build the assistant and as reference points for what "good" AI-assisted troubleshooting suggestions look like.
Screenshots
No screenshots are published for this project. The architecture and implementation notes above describe its current scope.
Lessons Learned
This is intentionally an experiment, not a production tool - the assistant's suggestions still need a DBA to validate them against the actual environment. The most useful part so far has been using it as a faster first pass over health check output, similar to how GitHub Copilot speeds up writing a script rather than replacing the judgment behind it.