Sublet.ai

Inspiration

Whenever you go onto the R/UCI subreddit or other similar university reddits, you see two sides of the same issue. People are looking for subleases, trying to find one that can fit their budget and needs, while not wanting to have to scroll through multiple posts or through other online websites to find one. On the other side, people post vague or lacking descriptions of their own sublease, which makes them potentially miss their target audience. This issue extends even outside of universities, to almost any Sublease situation. This is where our tool comes in, to solve both sides of the issue.

What it does

Sublet.ai allows users to upload a few photos of their own house or apartment and provide an address. This is all they need to create a very in-depth and detailed sublease posting. Our tool will use AI agents to analyze the photos and create a detailed description of how many rooms, bathrooms, appliances, and other amenities the house offers. It will then take the address and look around the posting. An agent will then take this address and search around the area for important information such as public transit options, information about nearby gyms, grocery stores, entertainment, safety problems in the area, and any other important information it can find through its web searching. This AI will take the core information that is needed for the posting and make it all available.

How we built it

We built the frontend with React and the backend with a mix of Node.js and Python. In the backend, we used lambdas to handle invoking and passing information between agents, AWS Bedrock agents to handle the image analysis and web searching, and API gateway for communication between the front and backend.

Challenges we ran into

We ran into many challenges throughout this competition! Setting up the communication between the frontend and backend to be able to send images was difficult. We had to figure out how to upload the images to S3 and associate them all under one job ID to then ask the backend to start only after they were finished uploading. Setting up the pipeline for agent-to-agent communication was difficult to design and implement. We had to learn about AWS Step Functions during this hackathon and create new Lambda functions for passing on data between agents and invoking them.

Accomplishments that we're proud of

Having set up a full pipeline that can analyze images and then take an address and use an agent to make its own web searches so that it can find important information around that address.

What we learned

We learned a lot about designing entire data pipelines through AWS services and the difficulties that come with setting up multi-agent workflows. We learned a lot about creating a proper web application with AWS services, setting up a backend with lambdas and API gateway, and using step functions to allow the backend lambda and services to talk with each other. Scoping and designing a full project and requirements was a very important lesson we learned, which will definitely help us to design better software in the future.

What's next for Sublet.ai

In the future, we would like to add another agent feature that we unfortunately had to cut due to time and feasibility constraints. Not only did we want a detailed posting description, but we wanted to take it to the next step and have an agent be capable of autonomously uploading that posting to sites like Craigslist, Reddit,Facebook, and other places where people usually look to find subleases.