greenflags
🏠 AI-supercharged roommate vibechecker. Winner of Best Design at SF Hacks 2024.
Overview
greenflags understands who you are, what you look for in a housemate, and is able to represent your deepest desires and qualities to deal with the roommate hunting conundrum.
Inspiration
Coming from halfway across the world, the first thing we had to solve before embarking on our journey in Silicon Valley was to first find somewhere to live. But San Francisco's skyrocketing housing prices meant we had to find roommates in a place we knew nothing about.
We've all heard the roommate horror stories on Reddit - 1 month into a 6-month lease only to find your roommate is a demon sent from hell! This is not only costly and time-consuming, but also potentially very dangerous.
Current solutions face several problems:
- Too focused on where you're living (how the house looks, what amenities there are, location) and not who you're living with.
- Even the most creative solutions ask you to fill up personality surveys and forms. But finding roommates isn't like grocery shopping, they're long commitments! We need a better way to express who we are and what we are looking for.
- Compatibility should go beyond numbers and capture the underlying subtleties in what we want - e.g cleanliness could mean "no dirty dishes in the sink" for one person and "not leaving dirty laundry around the house" to another.
Approach
- Matching people to people, not to homes
- Understand what you really want and who you really are
- Enhance the compatibility with AI
Features
greenflags moves away from conventional form-based questionnaires (e.g filling up forms) and instead capture the nuances in our responses by letting our users answer free-text responses.
Through the responses of each individual, the model can learn about you as a roommate and provide a grounded and objective representation of (1) yourself and your personality traits, and (2) your desired roommate attributes.
The platform will match you based on these differences, but also allow you to ask any leading prompt to your potential matches, to allow you to screen matches better.
Afterwards, you can then see the compatibility score of all your closest matches, a pentagon chart of the overlapping similarity traits, and an AI representation of their answers to your leading prompts since it is able to understand every user in the system.
Implementation
- Frontend was built using React, Bun and TailwindCSS
- Backend with Flask supported by FlagEmbedding's BAAI and OpenAI's GPT4 model
- Data is stored on MongoDB through Neurelo.
- Hosted on Google Cloud Platform