Inspiration
Real estate is all around us! It's an industry that makes a lot of money and is super influential in society.
Every month, renters have to pay rent. Renting kinda sucks and is not nearly as fun as owning a home. Don't you wish you had AI to help you find the perfect home to purchase? Don't you wish you had more space for your stuff or your family? Don't you have lots of questions about real estate?
Or maybe you have some real estate you want to sell. Don't you wish you had AI to help you sell your home? Do you have questions about selling a home?
What it does
User enters a question as a text query. For every input, the server makes at least four queries!
It makes at least two queries to Llama-index. It makes one query to Vectara. And one to Chat GPT Version 4.
By querying different AI tools, we can compare and contrast different results.
At the moment we have two custom data sets that the AI can retrieve from. One is a list of 400 houses and their prices in San Jose, as well as how many bedrooms or bathrooms, square feet, etc. Another is a list of 1000 cities in the USA and their average house price in that city over time.
How we built it
- Obtain data in CSV format and parse it
- Ingest our data into Vectara or load it into Llama Index Lambda Query Engines
- For a single query, find the results across all agents
- Display the results in a Flask Python Web server
Challenges we ran into
- A lot of our data is numerical! Dates, prices, size in square feet, number of bedrooms, number of bathrooms. LLMs are trained to be good at understanding words. But they cannot reason about numbers very well the way humans can.
To help us deal with this, we decided to use Llama Index's Pandas Query Engine. This converts the user's question into SQL-like Python code that gets executed on our server. The query can reason about numerical data and perform calculations about the data in the table.
Accomplishments that we're proud of
- For each user input, we try at least four different methods of getting an answer and show the results. This allows us to compare methods.
- We could easily add more methods and more data sources over time to make it smarter
- Did all of the work and real estate research during the one weekend event
What we learned
- How to ingest data into Vectara and make queries against it with the Vectara API
- How to load data into multiple instances of Llama Index's Pandas Query Engine and make queries against it
- Better understanding of real estate and how to get data about real estate
What's next for RARE: Really Amazing Real Estate
- Additional data sets to make it smarter
- Additional models to and tools to make queries and get results
- Rather than show all results from all tools, only show the best result from the different possible options
- For this weekend we focused on single family homes. But we could expand the project to condos or commercial real estate as well.
Log in or sign up for Devpost to join the conversation.