How Recruit Holdings, the company behind Indeed, taught its text-to-SQL agent to read a website's hidden context
Blog post from dbt
Recruit Holdings’ automotive marketplace team developed a Slack-based text-to-SQL analytics agent called Goro to help product managers and analysts query data, but found that conventional schema metadata, golden queries, and semantic models could not reliably distinguish among similar UX event logs. Rather than attempting to build a broad ontology, the team adopted an issue-driven approach that documents the hidden context behind specific failures in a YAML-based “web domain context” file linking user actions, interface elements, and generated log events. To prevent this context from becoming outdated amid frequent A/B tests and product changes, the YAML design file must be created and reviewed before features ship, while front-end logging code and QA tests are generated from the same specification. In a demonstrated case, this structured context enabled the agent to map a niche screen name to its correct event identifier and calculate an accurate unique-user count, and the team is now exploring how to capture user-journey intent so the agent can identify behaviors such as drop-offs rather than only count events.
No tracked trend matches for this post yet.
Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.