π About The Project
BlindAI is an open-source AI privacy solution designed to help individuals and companies query and deploy AI models while ensuring the privacy of their data. This is achieved via the use of a Python library that facilitates easy interaction with AI models without compromising data security.
The key feature of BlindAI is that it uses hardware-based Trusted Execution Environments (TEEs) to maintain the confidentiality of user data throughout the process. This approach ensures data and model protection, which is particularly important when dealing with sensitive information.
BlindAI is versatile and offers two main application scenarios:
- BlindAI API: This allows users to query popular AI models hosted by Mithril Security.
- BlindAI Core: This feature enables users to set up their own BlindAI server instance to securely host and deploy custom AI models.
Built With
BlindAI is built using several key technologies:
- Rust: A programming language promoting safe and concurrent systems programming.
- Python: A versatile programming language integral for the BlindAI library.
- Intel SGX: A set of security-related instruction codes that are helpful in ensuring data confidentiality.
- Tract: A rust-based library used for on-device machine learning inference.
These technologies form the backbone of BlindAI, contributing to its robust security and ease of use.
π Getting Started
For newcomers, it is recommended to begin with a Quick tour utilizing the open-source model Whisper, providing an introduction to the solution's functionalities.
BlindAI API Example
A simple illustration of how BlindAI API works involves transcribing an audio file:
transcript = blindai.api.Audio.transcribe(
file="patient_104678.wav"
)
print(transcript)
BlindAI Core Example
In this scenario, an AI company uploads and manages their AI models:
- Uploading and Deleting Models: An AI company can upload their AI model to the server, after which a unique model ID is assigned. This model ID can be used for future reference.
response = client_1.upload_model(model="./COVID-Net-CXR-2.onnx")
MODEL_ID = response.model_id
print(MODEL_ID)
- Client-side Model Execution: Clients can use the model to process confidential data securely, protecting it from unauthorized access.
pos_ret = client_2.run_model(MODEL_ID, positive)
print("Probability of Covid for positive image is", pos_ret.output[0].as_flat()[0][1])
π Getting Help
Users can seek assistance and support through various channels:
- Visit the Discord #support channel for community support.
- Report any issues by opening an issue on the BlindAI GitHub page.
- Schedule a meeting with the team by booking a session here.
π License
BlindAI is distributed under the Apache License, Version 2.0, which permits free use, distribution, and modification of the software. For more details, see the LICENSE.md.
π Contact
For further inquiries, Mithril Security can be contacted through various platforms:
- Email: [email protected]
More information about BlindAI is available on its GitHub project page.