Inspiration

Assignment briefs can combine deliverables, grading criteria, word counts, reference requirements, formatting rules, and submission instructions in dense blocks of text. Missing one requirement can create unnecessary confusion, even when a student understands the subject.

I built BriefTrace to help students turn those instructions into something easier to inspect and verify—not to write their assignments for them.

What it does

BriefTrace lets a student paste an assignment brief and analyze it with AI. It organizes extracted requirements into categories such as deliverables, grading criteria, word count, references, formatting, and submission instructions.

The key feature is traceability. Selecting a requirement takes the student back to its supporting passage in the original brief. BriefTrace also distinguishes explicit information from items needing review and details not found in the supplied text. It is designed to flag uncertainty rather than invent missing requirements.

How I built it

I used the Devpost Learn Skill Pack to follow a plan-first workflow. Through its guided process, I defined the problem and scope, documented the product requirements, and created a technical specification before moving into implementation.

I built the application with Bolt, React, TypeScript, Vite, and Tailwind CSS. A server-side Bolt Database Edge Function calls Google Gemini, keeping the API key out of the frontend. The analysis pipeline validates the structured AI response and checks each source passage against the original brief, allowing for harmless whitespace differences.

Challenges and what I learned

One challenge was keeping AI-generated requirements grounded in the supplied text. I added source-passage validation so that unverified passages are not presented as confirmed evidence. During testing, I also found and fixed a source-range mapping bug that would have prevented the highlighting interaction from working correctly.

Another challenge was API reliability. Gemini temporarily returned a 503 service-unavailable error, so I added limited retries with exponential backoff and clear error handling rather than pretending every request would succeed.

This project taught me to turn an idea into a scoped product, use planning documents to guide a coding agent, test both normal and failure cases, and verify important behavior instead of relying only on a successful build.

BriefTrace is intentionally focused: one pasted brief, a structured checklist, and evidence that students can inspect for themselves.

Built With

  • ai
  • bolt-database
  • bolt.new
  • devpost
  • education
  • google-gemini
  • learn
  • pack
  • react
  • skill
  • tailwind-css
  • typescript
  • vite
Share this project:

Updates

Submission history