Inspiration
The project idea came from an abandoned venture at s5 (nights and weekends, from buildspace.so)
Recently, reading https://news.ycombinator.com/item?id=44524544 got me interested again.
What it does
Allows the user to visually indicate when they want to exit local-only mode. At first, everything runs on the device with Ollama. Lots of people want this on their phones, but I don't want to use my phone! I want Jarvis, like in Tony Stark's lab. It definitely "looks" the part today.
How we built it
Python and Exa.ai to do the search and locally cache the research results for future queries.
Inference can be run via WandB, OpenAI, or Ollama (using llama3.3:70b by default)
Challenges we ran into
Scope creep. I started getting distracted by making it try to speak with Home Assistant.
Accomplishments that we're proud of
Got speech synthesis working locally with say command that comes with macOS.
What we learned
Voice assistants need stream=True, and all hardware needs simple bindings into C/Python.
What's next for Axel (Voice Assistant)
Integrating more hardware.
Built With
- exa.ai
- macos
- openai
- python
- wandb

Log in or sign up for Devpost to join the conversation.