Comfort Food Finder

Inspiration

I am travelling in Southeast Asia right now and have struggled to find restaurants that I liked. I realized there are certain restaurants and cuisines that don't fit my palette. I wanted to build a tool that learned from restaurants I know I like, learn what about those restaurants' dishes I enjoy (flavoring, spice level, etc.) and then recommends restaurants in the area I'm traveling in that fit my palette.

What it does

My project allows you to save restaurants you enjoyed, extrapolates necessary metadata from those restaurants and calls the Yelp AI API to find restaurants that best match your palette (e.g spicy level, types of cuisine, etc.)

How we built it

I used v0 + codex to built out the application and design the frontend components. I used Supabase to persist sessions.

Challenges we ran into

The Yelp AI API has a character limit of 1000 so I couldn't just dump a ton of context into the query. I wanted to use as much info (like reviews and a larger number of input restaurants), but had to summarize and truncate information that was passed in.

Accomplishments that we're proud of

I tested against restaurants in Seattle and it recommended restaurants I actually enjoyed!

What we learned

There's a lot that goes into building a good recommendation engine in that I have to make many tradeoffs in what I include and don't include into making those recommendations. The more verbose the features are going into the recommendation, the worse it makes. It's all about choosing the right inputs.

What's next for Comfort Food Finder

I would like to include Yelp reviews from restaurants into my query so that my comfort food search has more signals to use.

Built With

• nextjs
• supabase
• typescript
• v0
• yelp