Back to Projects

Dinner Match Lab · 2026 – present

A meal recommendation app that learns your taste from every swipe

I designed, built, and run Dinner Match Lab end to end: swipe-based recipe picks, a learned reranker, daily personalized top 10s, and the data pipelines behind them, on web and mobile.

Founder and sole engineer

Visit the live app

Personalized picks per user, daily

Top 10

Next.js and Expo apps

Web + mobile

Backend tests

180+

Problem

Deciding what to cook is a small decision that eats a surprising amount of time every day. Recipe sites give you endless lists; I wanted an app that gets to a good answer in a few swipes and gets better the more you use it.

Approach

  • Built a swipe interface where every like, dislike, save, and cook is logged as an event that feeds the recommender.
  • Designed a hybrid reranker: rule-based scoring for taste profile, cuisine, weekday, and holiday context, plus a scikit-learn model trained on each user’s recent interactions.
  • Generate a deterministic top 10 for each user every day, and keep anything recently seen or disliked out for 7 days.
  • Built ingestion pipelines that pull recipes and nutrition data from several sources, with parsing and data-quality fixes for ingredients, plurals, and fractions.
  • Ran it as a product: FastAPI and Postgres backend with Redis caching and scheduled jobs, daily email picks, a Next.js web app, and an Expo mobile app.
  • Replaced a paywall with optional Stripe donations to keep the core experience free.

Results

  • Live at dinnermatchlab.com, with a fresh personalized deck every day.
  • Recommendations adapt to each user’s swipes instead of a one-size-fits-all list.
  • Recommendation logic is covered by unit and regression tests, so ranking changes ship safely.

Stack

  • Python
  • FastAPI
  • PostgreSQL
  • Redis
  • scikit-learn
  • Next.js
  • React
  • TypeScript
  • Expo / React Native
  • Stripe
  • LangChain + Gemini