Learning path · 5 steps
Route written requests to the right team
People can describe the same problem in very different words, which makes routing a ticket harder than matching a single keyword. This path uses Jev to explore that decision with synthetic reports. You will define the queues, examine where its recommendations agree with your own and decide whether they justify connecting the model to an application.
In this guide 6 sections
Use a project you’re allowed to work on, and keep the first change easy to undo. Check what the tool proposes before applying it, and don’t send private data you aren’t allowed to share.
Explain what each destination handles
Start with the queues your example needs, such as access, application and network. Describe the reports that belong in each, and add triage for cases that need more information or do not fit. Jev’s Choice response selects from the labels you provide, so an unclear distinction between queues will also be difficult to interpret in its answers.
Include the reports that are hard to classify
Prepare synthetic tickets or approved sanitized text and label the expected destination before running the model. Include clear examples, overlapping symptoms and unfamiliar cases. When two reviewers disagree, examine the queue definitions or keep the ambiguity visible. Otherwise you risk judging the model against an answer the team itself cannot justify.
Compare the recommendations in the Playground
Use Choice in TypeSafe’s browser Playground. Keep each label, its distribution and the actual model version beside the original ticket. The distribution shows how the model weighs the available options; it gives you more to examine when a recommendation surprises you. For now, leave the results as recommendations so a wrong answer cannot send real work to the wrong team.
Compare with the routing you already have
Run the same reports through any rules or classifier already handling this job. Look separately at wrong destinations and cases sensibly left for triage, then include the human time required to correct or complete the routing. A predictable response format helps the software read an answer; the comparison tells you whether that answer helps the people receiving the work.
Connect the recommendation to delivery
If the recommendations are useful enough to integrate, look up the selected queue in the current destination directory and verify that delivery succeeds as a separate step. Before using confidence to decide which tickets can move automatically, compare the proposed threshold with reviewed outcomes. Keep the model version with the result so later changes can be checked against the same reports.
Tool guides for this path
An illustrative workflow based on the linked tool documentation. It has not been run as a product test or benchmark.
Browse all learning paths