unbored - Take the boredom out of onboarding

• 

  Our Team

Inspiration

New contributors routinely spend 2–4 weeks piecing together scattered READMEs, closed PRs, and tribal knowledge. We wanted a “first hour” experience that shows what this repo is, how it’s wired, and how to run it without multiple office hours or KT sessions

What it does

unbored turns any GitHub repository into a polished onboarding site in minutes:

• AI-generated architecture overview: summarizes services, modules, and data flow.
• Setup guide with pitfalls: pulls steps from scripts/manifests and mines historical issues/PRs for gotchas.
• File dependency graph: interactive view to trace imports and jump to source.
• FAQ from history: turns recurring questions into crisp answers with links.
• Zero-hassle hosting: outputs a static Docusaurus site you can drop on GH Pages/Netlify/Vercel.

How we built it

• CLI (Python + Typer) scans the repo: file tree, key configs, lockfiles, package metadata.
• GitHub analysis (PyGithub/GitPython) fetches issues/PRs to detect setup errors and repeated questions.
• Code parsing (Tree-sitter) extracts language-agnostic import edges (JS/TS/Vue + Python to start).
• LLM layer (Claude via API Gateway) generates architecture/write-ups, FAQs, and concise step lists.
• Site generator (Docusaurus + MDX) produces sections: Overview, Setup, FAQ, and a Graph View page

Challenges we ran into

• Graph layout quality: fcose on larger repos formed a big ellipse; we pivoted to precomputed layouts for stability.
• Packaging & distribution uncertainty: We weren’t sure whether to ship it as a hosted web app, a Python package/CLI, or a Node.js package

Accomplishments that we're proud of

• End-to-end pipeline in <24 hours from scan → summarize → graph → themed site.!!!

What we learned

Scope is everything. Focusing on a working MVP over perfection was key to finishing within the hackathon timeline. Iterate quickly. Short build-test cycles helped us identify what Claude did well and where it needed guardrails. Cross-disciplinary work pays off. Blending AI prompting, backend engineering, and UX led to a more complete product.

What's next for unbored - Take the boredom out of onboarding

• More efficient Claude Usage: Multi-prompt pipelines, token budgeting, and context recycling (because 200k tokens is still not enough for legacy code)
• Module level docs: Generate architecture + rationale per subsystem for complex and large databases
• Configurable Github and Claude API Access: Plug in your own GitHub token and Claude API key for private repos and custom rate limits

Built With

• amazon-web-services
• bedrock
• claude
• docusaurus
• html
• lambda
• node.js
• python
• react
• typescript