η.ai

η.ai showcases low-precision AI's speed and power efficiency, offering a free API for professionals. It explores analogue AI for even greater gains in efficiency. The source-code to this platform is provided in this repo.

Vision

The η.ai (eta.ai) project explores low-precision AI, challenging high-precision approaches. It reduces precision requirements, improving efficiency and hardware scalability. Low-precision AI (TinyML) swiftly analyzes large datasets in finance, cybersecurity, and healthcare. Affordable FPGA hardware, in this case Google Coral, supports low-precision AI, providing better performance and energy efficiency for edge computing.

How to use the service

You can create an account and login here: TBA

You can then use the API to send pictures to infer the objects that are in the picture. The callback is a JSON with the predictions and the power that was consumed to make this happen.

The interactive Swagger API documentation can be found when spinning up the service. The default address would be 127.0.0.1:8000. The OpenAPI documentation can then be found under at the /docs endpoint: 127.0.0. 1:8000/docs.

Do you wan to setup a service like this one for yourself?

Then you can! Please note that this platform was implemented using:

• Linux (debian)
• Docker (you can deploy instantly) and docker tools like docker compose
• For API-handling FastAPI was used, a python framework
• Python dependencies are managed using Poetry

You can run the code directly on your machine. In that case you need to:

1. Clone the Repo
2. Install dependencies using Poetry and drivers using the systemdeps/install_and_run.sh script. This script was install, test and run the API.
3. For running the api seperately, you can use poetry run uvicorn main:app --reload
4. You can find the swagger documentation ip-address:port/docs

If you want to run the docker-container, please make sure that you:

1. Install the driver on your OS first using systemdeps/install_drivers.sh
2. run docker-compose up
3. You can find the swagger documentation ip-address:port/docs

Note that the docker-compose approach adds additional services and is intended to be used as

License

The license can be found in the UNLICENSE file. For more information, please go to: unlicense.org. If you want to have more information on the licensing topics as a whole, please checkout choose a license.