Know when the right decision changes.

Inspiration

A small change in circumstances can turn a previously correct decision into a mistake. A trainee can remember the usual steps and still miss the exception that changes what to do. CRIVANT helps people practise recognising those changes before they encounter them at work. It gives training coordinators and supervisors a short, procedure-based rehearsal and a clear idea of what to revisit.

What it does

A facilitator approves three cases based on the same rule. In the first, the usual conditions apply. In the second, one fact changes and the correct decision changes with it. In the third, a detail changes and the decision stays the same. The trainee answers each case before seeing feedback. CRIVANT cites the rule behind the review and returns Cannot grade when the required evidence is missing or conflicting.

In a fictional card-routing drill, a missing review seal changes where a card should go. Sleeve colour has no effect. One trainee routes correctly at first, keeps routing when the seal is missing, then holds the card when only the sleeve colour changes. Another holds every card. Both score one out of three. The first trainee can practise spotting the missing seal and disregarding the sleeve colour. The second needs to revisit the baseline rule.

How we built it

The Skill is designed around the organisation’s own procedure stored on UGREEN AI NAS. It uses authorised NAS access to locate the document, retrieve the relevant passages and record a file fingerprint. When a facilitator reopens a saved receipt, the Skill is designed to flag changes to the source file for review. With approval, the Skill can save a training receipt to an authorised NAS folder and check that it was written. Confidential material requires verified, organisation-approved local processing.

Challenges we ran into

Each case needs a decision that follows from the approved rule. We built the Round 1 Uliya Skill package to draft cases from an identified rule, require facilitator approval, record each response and give feedback backed by the procedure. When a required passage cannot be verified, the Skill holds the exercise for review.

Accomplishments that we’re proud of

The official Uliya checker returned zero errors and warnings. Nine original deterministic fixture tests and eight targeted contract checks passed locally using fictional documents and structured inputs.

What we learned

The fictional drill shows how the same score can call for different coaching. Checking why a choice changed can be as useful as checking whether it was correct.

What's next for CRIVANT

Live model behaviour, local model routing and NAS execution still await device validation.

Workplace training is the first use case. The broader value is giving people room to make a decision, explain it and get specific feedback before they face the same choice on the job. Future pilots could explore instructor-led apprenticeships, vocational classrooms or public-service onboarding wherever people work from approved procedures. CRIVANT is a rehearsal aid. Qualified instructors and the organisation’s approved procedures remain the authority for real work.

Built With

  • ai
  • document-processing
  • file-management
  • markdown
  • openai
  • ugos
  • ugreen
  • uliya
Share this project:

Updates

Submission history