Llamour

Inspiration

We were annoyed from the repetitiveness of dating apps. We had this idea of bots doing the annoying part where we could just start talking when the discussion becomes interesting.

What it does

Llamour inspires itself from your talking manners and hobbies to create discussions with the AIs of other people. Just start to talk when you think the discussion gets interesting.

How we built it

Built with a lot of amour... Llamour uses a Python server connected with a Flutter application for allowing the user to interact. It uses VAPI embedded with Flutter for the Speech-to-Speech part, MistralAI API for extracting of information from the people's discussion, judging the perfect match from the resulting discussions ran earlier with Groq's API.

Challenges we ran into

Biggest challenge, Mistrals models tend to deviate from the given instructions (comes back to being an assistant instead of the roleplay of the user profile)
Instability of Speech to Speech pipeline with model's answer coming from LLM inference.
Making a native mobile app in 24 hours

Accomplishments that we're proud of

A smooth experience with Speech-to-Speech, a smooth and minimalistic UI in 24 hours, faster conversation generation, fullstack infrastructure, models fidelity to discussion informations passed inside the Llamour.

What we learned

To sleep or not to sleep during a hackathon.

What's next for Llamour

Vector similarity search, recommendation algorithm, a personalization's process for the user, Speech-to-Speech between clones, a LVM for understanding images in the conversation and sending pictures (allowed by the user) between clones.