Team 22

MealMatch

MealMatch is a personal agent that simplifies the food-ordering user experience from deciding the meal to checkout, automatically personalizing recommendations based on price, transaction history, and user preferences across delivery apps like Uber Eats and DoorDash.

Check your own project

Community votes recognize the build and do not affect official judging.

Review the project

Start with the source code, then open the demo or video if available.

View GitHub repository
Project description
Our team of 3 built MealMatch, a personal food-ordering agent that compares food delivery apps like Uber Eats and DoorDash, remembers your taste across platforms, and backs every recommendation with an audit trail. A LangGraph agent loads your memory, parses the request, searches both apps, matches the same dish across them, filters by your rules and budget, ranks, and explains its top 3. It incorporates the human in the loop twice: when the user chooses a meal, and when the user decides to approve or reject the order. Every step logs evidence, inference, action or approval, and every claim links to the fact behind it. Superlinked SIE runs all model work. For example, Qwen3.8 27B parses the human request and explains, R3 embeddings matches dishes across apps, and Qwen3 Reranker 4B ranks dishes against your request and taste. SingleStore stores memory, dish embeddings, and the full audit trail. Cognee turns past orders into learned preferences that cite their source orders. Then, Statsig gates which ranking algorithm each user gets and logs every step as an event. Finally, Firedrill runs 10 drills against a simulated shop, including prompt injection in menus, to prove it never checks out without a yes.
Tools used