Leasa | Devpost

Inspiration

Tired of them ghosting you, charging a fortune, or just existing? So were we. That’s why we made Leasa—your AI-powered rental sidekick that kicks brokers to the curb. Leasa connects landlords and tenants directly, no commissions, no shady middlemen, just pure unfiltered apartment bliss. Smarter matches, fewer headaches, and absolutely zero awkward “just checking in” texts from Chad the broker.

What it does

Leasa is an AI-powered real estate assistant that:

✅ Connects tenants and landlords by understanding both sides' preferences through conversation ✅ Acts like a real agent — chatting naturally, answering questions, and guiding users ✅ Recommends properties or tenants that best fit what the user is looking for ✅ Stores chat history, so every interaction feels personalized and continuous ✅ Simplifies property management, letting landlords upload listings and get matched instantly

Think of Leasa as your smart, always-on leasing agent — without the back-and-forth, missed messages, or guesswork.

How We Built it

We used Gemini 2.5 for the agent features. For this low-fidelity prototype, we decided to do everything locally so chats were stored in a json file. This allowed us to focus more on other areas like model improvement. Used FastAPI for the backend.

Challenges we ran into

Using Gemini to do our matching instead of a dedicated matching algo. We also ran into issues where the result would be out of someone's price range or location preference, 2 non-negotiable factors in finding an apartment. We fixed the issue as it was caused by the agent not having the ability to return no listings.

Accomplishments that we're proud of

We're proud of making a product that we believe has real world value, and especially one that solves an issue experienced by hundreds of millions of people. As students, we know how hard it is to search for an apartment, and we are excited at taking a shot at addressing that for fellow students.

What we learned

Finetuning a model to follow the correct protocols involves creative ways to structure the output. Structured outputs itself is a discipline that we each learned a bit of in this time. Also, learning how to work with voice capabilities and Google ADK were great learning moments.