Inspiration

I've always been interested in developing an AI or "intelligent" companion app that is multimodal and human-like in terms of responses.

What it does

It is an electron app that makes it easy to transcribe a user's voice and respond via voice. The companion is personalized to use Google Gemini's pro tts model and the voice Leda for now but this is easily modifiable.

How we built it.

The app is bundled using electron-forge to be a cross platform app in the future and uses svelte to render the UI.

Challenges we ran into

Some of the challenges I ran into was trying to reduce the latency of the response from the google api, since the entire app is modular, the responses from each API is a massive bottleneck in trying to make the app responsive. Getting the "companion" to remember and implement a bandaid for chat memory was also one of the biggest challenges.

Accomplishments that we're proud of

I'm proud that I managed to implement a companion like app from scratch.

What we learned

What's next for Project-EVA

Screen capture, image & screenshot recognition, refractoring the codebase from Electron to Tauri (Rust-based) so package binaries will be smaller and the app will be more lightweight so that it can run properly on lower end computers.

Built With

Share this project:

Updates