← Back to projects
03Learning · AI

Vantage

An AI learning app that trains client-facing consultants to judge which AI approach fits a client problem.

Live siteRead the PRD
Vantage login page
88%of testers finished a meaningful activity with no help (15 of 17)
12 / 14solutions rated helpful
268unit tests

Roadmap, from the PRD

V1 · Shipped
  • Guess-then-reveal practice loop
  • 90 real-work scenarios and 120 concept questions
  • The reveal explains why each alternative loses
  • Ask a real client problem, get a step-by-step solution
  • First-guess accuracy over time, and a history of every attempt
  • Private by default: no streaks, points or leaderboards
  • Content pipeline with a human review gate
  • Admin analytics dashboard
V1.1 · Next
  • Automatic question generation and approvalReview by hand does not scale past a pilot.
  • Lightweight rating on the reveal stepCloses the rating-coverage gap.
V1.2 · Later
  • Skill Improvement Rate scoringIt needs a validated rubric first.
Cut
  • Share and inviteNo evidence earned it, and it conflicts with the privacy principle.

User flow, from the PRD

Vantage user flow diagram. Solve answers a real client problem. Practice builds judgment with a guess-then-reveal loop. Progress shows accuracy, history and strength by category.
Two surfaces, one judgment engine. Solve answers the problem in front of her. Practice builds the judgment that means she needs help less next time. Click the picture to open it full size.

Challenge

Professionals in India use AI every week, but cannot tell which AI approach fits a new client problem. Courses add more content, and free content is completed by fewer than 5% of learners.

Approach

  • I ran 9 interviews across 8 roles before choosing a persona. The client-facing consultant came out on top: the problem comes back every week, a wrong approach reaches a paying client, and the answer can be graded.
  • Guess before reveal: the answer shows only after the learner commits, and the reveal explains why each other option loses
  • 90 scenarios and 120 concept questions, in difficulty tiers
  • Solve: bring a real client problem, get a step-by-step approach
  • A human review gate: no AI-made question reaches a learner unchecked

Result

17 external testers used it with no onboarding call. 15 of 17 finished a meaningful activity with no help, and 12 of 14 rated the solutions helpful. Only 2 of 17 found the scenario quiz, so I changed the navigation. I cut share and invite, streaks and payments, because no evidence supported them.

Case study
Rethink Systems: Learning tech and AI
Role
Interviews, PRD, design, build, pilot
Built with
Next.js, TypeScript, Supabase, Groq, Vercel