Skip to content

Latest commit

 

History

History
25 lines (14 loc) · 1.65 KB

File metadata and controls

25 lines (14 loc) · 1.65 KB

Build Log — AI Usage

This log documents where AI (Claude) assisted in building this capstone, what it got right, what it got wrong, and what I changed. Kept honest and updated as I build, per the capstone's ground rules.

Architecture & stack decisions

  • Initial stack (Supabase) was replaced with Neon Postgres + Prisma + Auth.js after I confirmed I could only use MongoDB or Neon — documented here because it changed the auth model (Supabase Auth → NextAuth v5 credentials + bcrypt) and removed an entire folder (supabase/) that had already been scaffolded.
  • Chose Upstash Redis for rate limiting specifically because in-memory rate limiting (e.g. plain express-rate-limit) does not work correctly on Vercel's stateless, multi-instance serverless functions — this was flagged by Claude, not something I already knew, and I verified it against [reasoning/source].

Where AI helped

  • (fill in per file/feature as you review and understand it — e.g. "geo fallback chain in enrich.ts: understood and can explain the try/catch loop and why it never throws")

Where AI was wrong / had to be corrected

  • lucide-react was initially installed as ^1.28.0, which doesn't match that package's real versioning (it's on the 0.x line) — caught and corrected to ^0.462.0 in package.json.
  • (add any other corrections you make while testing)

What I changed after generation

  • (e.g. any UI tweaks, renamed fields, adjusted rate limit thresholds after testing)

Lines I can explain unprompted

  • (before your demo, pick 5–10 specific lines across different files and make sure you can explain them without looking — this is literally what gets tested)