StudyDebt

StudyDebt AI — Discover the hidden cost of skipped learning.

Comment

Devpost Submission — StudyDebt AI

Inspiration

Every student has skipped a lecture or a topic. Most think it's fine — they'll catch up later.

They don't realize that skipping Recursion doesn't just mean they don't know Recursion. It means they also can't learn DFS, Backtracking, Dynamic Programming, or Segment Trees. One skipped topic silently breaks an entire branch of their education.

StudyDebt was built to make that invisible damage visible.

What it does

StudyDebt AI is a 7-step agentic AI pipeline. You enter a subject and the topics you skipped. The agent does the rest:

  1. Knowledge Auditor — Scores your knowledge debt from 1–100 and estimates how many days it will take to recover
  2. Dependency Mapper — Identifies every topic in the course that directly depends on what you skipped
  3. Future Failure Simulator — Picks your most critical broken dependency and simulates you attempting it step-by-step, revealing exactly which step fails and why
  4. Consequence Predictor — Predicts your interview risk, exam risk, and long-term career impact
  5. Recovery Planner — Generates a personalized day-by-day recovery roadmap with curated resources
  6. Quiz Generator — Tests your weakest gap with a targeted diagnostic question
  7. Adaptive Recovery Agent — Analyzes your quiz result and rebuilds your recovery plan in real time — simplified if you failed, accelerated if you passed

Each agent receives structured output from the previous step, forming a stateful reasoning chain.

How I built it

  • Frontend: Vanilla HTML, CSS, JavaScript — no frameworks, no build tools
  • AI: Groq API with LLaMA 3.3 70B for all 7 agent steps
  • Architecture: Each agent call passes structured JSON context to the next, creating a dependency chain rather than isolated prompts
  • Deployment: Vercel

The entire pipeline runs client-side. No backend, no database, no infrastructure.

Challenges I ran into

  • Getting all 7 LLM calls to return strict JSON reliably required careful prompt engineering and robust parsing
  • Designing the Future Failure Simulator to produce believable, specific step-by-step failure traces took multiple iterations
  • Making the Adaptive Recovery Agent genuinely respond differently to pass vs fail required testing many prompt variations

Accomplishments that I'm proud of

  • A complete 7-step agentic pipeline built solo in under 10 hours
  • The Future Failure Simulator — showing exactly where a student will fail before they get there
  • The feedback loop: quiz → adaptive plan. This is what separates it from a simple prompt chain
  • Clean, polished UI that makes the agent pipeline feel alive

What I learned

  • Agentic AI is about context passing, not just chaining prompts
  • The most impactful AI products reveal something people didn't know about themselves
  • Constraints (10 hours, solo, no backend) force better product decisions

What's next for StudyDebt AI

  • Support for uploading a syllabus PDF to auto-detect skipped topics
  • Persistent progress tracking across sessions
  • Spaced repetition integration with the recovery roadmap
  • Multi-subject debt tracking dashboard

Links

Built With

groq llama-3.3-70b javascript html css vercel

Built With

Share this project:

Updates