Side project · Live

Saveur AI

Group dinner planning: Claude recommends restaurants for everyone's tastes, and the group votes. Built for fun.

What it is

Group dinner planning with AI-powered restaurant recommendations. Members share their preferences (budget, cuisines, dietary restrictions, vibes), Claude suggests restaurants that suit the whole group, and everyone votes. I built it for fun, and to keep my hands on the model APIs outside of work.

How it works

One prompt is built from every member's profile and returns structured JSON: 20 restaurants, each with a reason it suits the group. Real venues come from the Google Places API.

Engineering notes

Recommendations and place searches are cached for 24 hours and capped per user per day to control cost. Model errors and malformed output return explicit errors instead of failing silently. The API uses Supabase auth, per-route rate limits, and a restricted CORS policy.

Stack: FastAPI, SQLAlchemy, PostgreSQL (Supabase), Next.js, Anthropic API.