i decided to pause this Git-Repo till i got an alpha version done. The initial goal of showing my project concept to a jury of the german goverment was succsessfully compleded. I am currently working on a more complete Library. I aim to come back with a publishable version before the last quarter of this year! Untill then feel free to have a look arround and see what the concept is all about.
At Use-AI.rs, we are building an open-source AI framework in Rust. Our goal is to create a concurrent, locally hostable AI agent for practical applications in production environments.
To achieve this goal, we are building on top of Burn and CubeCL. We are working with a JSON configuration standard. This configuration file contains all the information required to transform data and execute various AI operations. CubeCL provides us with the necessary operational tools to achieve our aim of delivering a practical AI framework.
For production coding, we will focus on implementing reinforcement learning (RL) algorithms such as Q-Learning and Deep Q-Networks. Additionally, we plan to implement ensemble learning algorithms. In the first iteration, we will use gradient-based decision trees (GBDT).
To provide insights into our decision-making process and document early-stage changes, the maintainer is maintaining a small blog on The Hitchhiker's Guide to Use-AI.rs.
This tool represents the highest level of abstraction we provide, in the form of a CLI tool. The CLI tool will include:
- A configuration manager and
- A staging tool for exploring the features of Use-AI.rs with custom data.
- No prior Rust knowledge required: Users can explore the tool without needing to understand Rust.
- Comprehensive implementation guidance: The tool's code includes detailed explanations for production use.
- Prerequisites: Rust and MongoDB must be installed.
This tool is built using Inquire.
The Store component provides a simple synchronous MangoDB storage layer for:
- Storing configurations.
- Storing outputs.
The stored data can be downstreamed for further use.
The Stage component processes and prepares abstract information for the Operator. Key features:
- Prepares data for downstream processing.
- Parallelizes some operations using Rayon on the CPU.
Future Plans:
- High-Performance Computing (HPC): Enhance parallelism and optimization for large-scale data streams.
The Calculator component processes prepared data into Burn Tensors. Key features:
- Executes Tensors using the WGPU backend for now.
- Designed with reinforcement learning (RL) in mind.
Future Plans:
- Build custom GPU kernels using CubeCL for more complex RL operations.
