Introduction

Overview

Building robust models from data requires close collaboration between domain experts and data scientists. Once the algorithm is trained, the process becomes similar to standard engineering activities.

AI Problematic

AGILab is a multi-project data science solution designed to manage projects using template-based frameworks. It also serves as a first step towards an experimental Python IDE for data science applied to engineering.

Remark: Historically, AGILab was developed as a playground for agi-core. If you only need cluster automation, you can use agi-core directly (for example from a notebook) without the Streamlit user interface.

Purpose

AGILab aims to enhance data science productivity

  • Cross-Functional Collaboration in Data Science: Provide project templates plus import/export and cloning workflows to promote code reuse across teams.

  • Enhance Agility in AI Projects: Provide scalability for both data and execution so you do not need to rework your code as datasets grow. Provide built-in online and offline generative AI helpers to assist algorithm exploration and coding assistance (prompt engineering).

  • Overcome IT Constraints: Automate deployment on local or SSH clusters so you can run workloads at scale without bespoke DevOps workflows.

Target Audience

AGILab is intended for professionals and domain experts, offering a tool that simplifies AI project development.

Main Dependencies

  • uv: Advanced utilities to manage Python virtual environments.

  • asyncssh: Enabling SSHv2 protocol on top of the Python asyncio framework.

  • Cython: Cython is an optimising static compiler for both the Python programming language and the extended Cython programming language (based on Pyrex). It makes writing C extensions for Python as easy as Python itself.

  • Dask: Parallel computing for handling large datasets and computations.

  • Streamlit: Rapidly develop and deploy interactive applications.

  • Pycharm: Assist you for advanced code development with AGI.

  • OpenAI: Assist you for algorithm discovery and code generation.

  • Mistral (Universal Offline AI Chatbot): Provides local Mistral-instruct models for offline experimentation.

  • GPT-OSS Offline: Runs open-weight GPT-OSS responses locally so experimentation keeps working without external connectivity. Configure the assistant via the Streamlit sidebar or the GPT_OSS_ENDPOINT and GPT_OSS_MODEL environment variables (defaults to http://127.0.0.1:8000/v1/responses and gpt-oss-120b).

Note

Windows builds are still catching up. Features that rely on local assistants—such as the GPT-OSS offline coding helper—are only partially supported while the port is in progress.

Technologies Selection Criteria

The selection of these technologies have been done for Portability, Simplicity and Performances:

  • Portability: AGILab runs and shares environments across systems without the overhead of full virtualization.

  • Simplicity: AGILab manages dependencies and isolates project environments, reducing conflicts.

  • Performance: AGILab provides runtime benchmarks for combinations of Cython, Dask, and thread/process execution modes.

Example: Aircraft Radio Communication

  • Needs Algos Set

Today, a single algorithm is rarely enough to solve complex problems. Instead, a suite of algorithms is needed to address diverse use cases with intricate implementations. For example, consider telecom systems in an aircraft:

Needs for Algo Set

This scenario requires reuse across organizations, infrastructure, and operating systems. While many consider virtualization and containerization (e.g., Docker) as the solution, AGILab offers an alternative.

AGILab is designed for modern development needs—reducing overhead, promoting collaboration, and simplifying the deployment pipeline without compromising on power or flexibility.

  • Needs Model of Models

Since Artificial General Intelligence (AGI) has not yet been achieved with a single algorithm, a “model of models” is necessary. For instance, in aircraft radio communication, this approach is illustrated as follows:

Model Of Models

AGILab provides a dynamic playground to experiment with various architectural solutions without reworking your code. This flexibility is achieved through two main mechanisms:

  • A versatile GUI that allows selection from 16 different execution modes.

  • A decoupling of software architecture from available hardware resources.