Modern LLMs
Modern LLMs are now capable of generating python code to temporarily train and test models in the chat.
Core trick
MCP server lets Le Chat train models on uploaded datasets with tuned hyperparameters, persisting models for reuse across chats.
Description
Modern LLMs are now capable of generating python code to temporarily train and test models in the chat. Our connector makes everything persistent and manageable
The Predictif MCP connector Lets you manage models and datasets right in your chat. Train model on your data with custom parameters, and make your models persist on the MCP server.
Reuse models and datasets in new conversations, and run predictions on fresh inputs. You can also save trained models to LeChat’s library for easy sharing and downloading.
README excerpt
# Prédictif  ## About Prédictif is a MCP Server for training ML models in Le Chat. It let you : 1. chooses optimal model for your data 2. Optimizes model parameters without manual intervention 3. Handles preprocessing, training, and validation 4. Exploit the data and give back insight to the user The project is conducted by the team _HireUsFaang_ composed of - x_13a (Alex Foulon) - slavastar7 (Vyacheslav Efimov) - alexvplle (Alexis Vapaille) - dirdros (Adrien Pelfresne) ## Installation ```sh cd predictif uv sync uv run python -m predictif.main ``` ## Documentation [Documentation](./docs/) is available in markdown format