homefinder

logo

Inspiration

We identified a gap in the market where numerous housing databases lacked an efficient and adaptable user interface. This triggered the idea for homefinder - an application designed to bridge this gap and maximize the potential of housing databases.

What it does

homefinder is a react frontend and flask backend which takes user prompts, enriches the prompt with database information, and sends it to ChatGPT. ChatGPT is able to pass queries to the database to get information it needs to send to the user. Then, along with a response from ChatGPT, a map is displayed, with the locations.

homefinder's scalability is attributed to its decoupled architecture with a React frontend and Flask backend, allowing for independent scalability of each component based on demand. Its integration with ChatGPT ensures quick and consistent responses, while its backend communicates effectively with any housing database.

Its horizontal scalability ensures high performance, as more instances can be added as traffic increases. The system's adaptability extends to compatibility with any database schema or data type, contributing to its seamless integration. Its main scalability limitation lies in the database interaction, which can be managed effectively with proper database optimization, making homefinder a scalable and efficient solution for diverse housing databases and user demands.

How we built it

We built it through teamwork and the bringing together of experts across different disciplines. Our front end is a simple interface for our back end and, we built an interface over a database, which can be adapted to any database. By doing this, we are able to then adapt our database, or even use different databases.

Challenges we ran into

Working together can be difficult. We were able to overcome some of the challenges in working together by establishing roles early on, creating the journey, and aligning contracts between the various moving pieces.

We also ran into prompt-engineering issues. Prompts that are too large, because we have such specific needs.

Accomplishments that we're proud of

We had coded a lot of flows into our backend which we were able to remove and simplify, using ChatGPT. For example rather than provide specific queries to our database, we passed the schema to ChatGPT and let it decide how to query the database.

We are particularly proud of how homefinder adapts to any housing database, regardless of its structure. The flexible and adaptable nature of the LLMs, which ensures seamless interactions and accurate results, is a significant accomplishment for our team.

What we learned

The development of homefinder has taught us the immense potential and versatility of language learning models. We also learned how to effectively deal with a wide variety of data sets and schemas, and how to create a flexible user interface that can adapt to any database.

We also learned that having such a rigid structure for the response and workflow was not great, and we want to make it as flexible as possible to leverage LLMs.