Inspiration
The project was inspired by a limitation we noticed while working with Yelp’s Conversational AI API. The API does not store personalized user information, which meant users had to repeatedly provide their location, work or school address, dietary restrictions, favorite cuisines, and other preferences in every conversation. This made interactions longer, repetitive, and less effective. We wanted to solve this by building a platform that could persist relevant user context and preferences and inject that information into the initial AI request, resulting in shorter, more precise, and personalized conversations.
What We Learned
Through this project, we learned how to design a system that maintains conversational context across sessions, how to structure and optimize AI request payloads, and how frontend and backend coordination impacts performance and user experience.
How We Built It
We built YelpMe using a React frontend powered by Vite, with a Django backend and Django Ninja for a fast, typed API. User data and preferences are stored in a SQLite database, and the application is containerized with Docker for easy setup and deployment.
Challenges
One of the biggest challenges was optimizing API response speed. We overcame this by sending all relevant user data in the initial request and then using the chat ID returned by the API to maintain the same conversation context for subsequent messages. This avoided resending large payloads on every request while keeping the interaction fast and consistent.
Accomplishments that we're proud of
We successfully built a fully functional, end to end discovery platform that adds a personalization layer on top of Yelp’s Conversational AI API. We designed a system that persists user preferences and chat context, allowing conversations to be shorter, more precise, and more useful. We are also proud of the clean separation between frontend and backend, the fast and typed API built with Django Ninja, and the performance improvements achieved by optimizing how and when data is sent to the AI API.
What we learned
We learned how to design and implement stateful AI driven conversations, how to efficiently manage and inject user context into AI requests, and how to optimize request payloads for performance. We also gained experience building a modern full stack application, coordinating frontend and backend development, and integrating a third party AI service in a reliable and scalable way.
What's next for YelpMe
Next, we want to expand YelpMe beyond discovery into action. A key feature we plan to implement is calendar scheduling, including integration with third party calendars. Using availability information returned by the AI API for restaurants or services, YelpMe would be able to suggest and schedule reservations or appointments directly on a user’s calendar. This would allow users not only to find the right place, but also to seamlessly book and plan without leaving the platform.
Log in or sign up for Devpost to join the conversation.