AI Apartment Concierge

Inspiration

Our project was inspired by the challenges of finding an apartment in a desirable location.

What it does

When you send a message on the chatbot, it searches for apartments in the specified area and provides the best results.

How we built it

AI Apartment Concierge leverages the Cloud3 API to parse and generate optimal results for the user. We utilized AWS, Lambda, and LaunchDarkly to ensure efficiency and reliability.

Technical Details

Cloud3 API: Parses and generates results.
AWS Lambda: Handles serverless computing tasks.
LaunchDarkly: Manages feature flags for efficient updates.

We also used high-powered language models (LMs) to embed messages into vectors that represent their meaning. Thinkmap.ai was instrumental in planning our direction and iterative development.

Our MVP included a tool for thesis statement and essay analysis, colorizing sentences based on their relevance to the thesis statement. We brainstormed ways to leverage underlying vector encodings to add depth to our text analysis. Convex was used for chat synchronization, dynamically computing embeddings and cosine similarity without issues.

Challenges we ran into

We encountered several challenges:

Ensuring everyone was connected.
Debugging 500 errors.
Issues with Lambda connectivity.
Internet lag due to high usage.

Despite these obstacles, we successfully completed a working project within a 6-hour timeframe, which we consider a significant achievement.

Accomplishments that we're proud of

We formed a team and shipped an innovative AI product in just one day during an in-person hackathon.

What we learned

The tools used during development can significantly reduce data management headaches.
Working with LM embeddings is more cost-effective and meaningful than pure text generation.

What's next for AI Apartment Concierge

Future plans include:

Adding an avatar to dictate results and interact with users.
Optimizing messages and displaying direct links to apartment listings.
Listing up to three apartments or locations that meet user criteria, complete with images.