How We Built AI-Powered Triage for Closot Requests

AI
Nikita Rani·

When a new request arrives in Closot, the triage system analyzes it, assigns a priority, suggests a team, and adds labels — all within 2 seconds. Here's how it works.

Context injection

The triage model doesn't just look at the request text. It pulls context from the workspace: the requester's role and team, similar past requests and their resolutions, the current workload of potential assignee teams, and any related ongoing projects. This context transforms a generic "something is broken" into a properly labeled, correctly routed ticket.

Triage pipelineNew requestContext injectClassifyRouteWorkspace context: requester role, past requests,team workload, related projects, similar tickets

Transparent decisions

Every automated decision includes a brief explanation: "Routed to Platform team because 3 similar issues were resolved by them in the last month." Users can override any recommendation with a single click, and overrides feed back into the model as training signal.

Since launching triage, average time-to-assignment dropped from 4.2 hours to 12 minutes, and mis-routing decreased by 73%.

Nikita Rani·
Copy link