projects / larpedin
- year
- 2026
- status
- live
- stack
- Next.js, TypeScript, Tailwind, Supabase, Postgres, Gemini, Vercel
Import your profile, get an AI career rating, then put it up against other people in 1v1 ranked battles. Win and climb from Bronze III to Grandmaster, with an AI coach telling you what to fix next.
Profiles are scored across seven weighted dimensions - experience, projects, technical, education, leadership, communication, and profile quality - so the number is harder to game than a single score. Résumé PDFs go straight to Gemini rather than through text extraction, which handles two-column layouts and scans that extraction mangles. Files are parsed and discarded, never stored, and nothing saves without review. If the LLM is unavailable, a deterministic heuristic engine takes over, so the app is never down because a provider is.
Battles run on standard chess Elo with rank tiers on top, plus XP, levels, daily streaks, achievements, and global, school, and country leaderboards. An anonymous mode hides name, headline, company, and school so profiles get judged on substance.
Built on Next.js and Supabase. Every gameplay mutation goes through SECURITY DEFINER functions instead of table writes, so clients can't set their own Elo or XP - battle resolution rejects judging a battle you're in, or naming a winner who isn't one of the two combatants, at the database level. RLS is on every table.
Deployed on Vercel.