🏠 studentHomes

An intelligence solution to finding student accommodation.

💡 Inspiration

Finding student housing whilst balancing your studies can be a length process, and most of the workload comes from having to find houses that you are interested in, and filter out those that your are not. Our project studentHomes aims to allow students to intelligently browse housing options, making it easier than ever to find their next home.

⚙️ What it does

• Learning via Human Feedback: Users can calibrate their preferences by training a model through small sample preference analysis, the more the user provides feedback, the more accurate their recommendations.
• Smart Search: Using semantics and embedding, users can intelligently search for their ideal home using natural language prompting.
• Photo Analysis: Arbitrary ratings are applied to housing options based on the provided photo for the property.

🛠️ How we built it

🧠 Challenges we ran into

One of the challenges that we faced was when trying to produce usable distances for our embeddings for the details about properties, this was the backbone of the core functionality of the semantics searching for properties and providing relevant results. Another challenge that we faced was getting a suitable dataset of property images that we could train our photo analysis model on, most of the datasets we found were not relevant or did not have a large enough volume of images, so we had to scrape existing properties from services like UniHomes.

🏆 Accomplishments that we're proud of

This was our first experience with trying to move away from heavily relying on LLM API calls to build out the core functionality of the project, and instead went a new direction with trying to build intelligence features with some machine learning ideas and techniques in mind from the ground up. This was something we had never tried before and we were proud to have been able to implement these features.

📖 What we learned

We learned about how to leverage human feedback whilst training models for use in recommendation systems. We learned about how this method is implemented in more depth with RLHF in large scale production products such as OpenAI ChatGPT and how we can use these techniques in our own projects.

Built With

• chromadb
• fastapi
• openclip
• python
• pytorch
• react
• sentence-transformer
• sqlite
• tailwindcss