Logo Lanfrica

Lynchpin707/Mentorati

Domaine:

natural language processing

Type de record:

software
Créateur:
Lyn
Hôte:
Mentorati is a local, RAG-based (Retrieval-Augmented Generation) Spanish Grammar Tutor. The name is a mix of the word "Moualimati", which means my teacher in the morrocan dialect Darija and the Spanish word "Mentora" which stands for mentor/teacher. # Mentorati (منطورتي) **Mentorati** is a local, RAG-based (Retrieval-Augmented Generation) Spanish Grammar Tutor. The name is a mix of the word "Moualimati", which means my teacher in the morrocan dialect Darija and the Spanish word "Mentora" which stands for mentor/teacher. Mentorati leverages local LLMs and a custom pipeline to provide explainable, context-aware grammar assistance without sending data to the cloud. Very usefull for when you don't have Wi-Fi but still need to study some Spanish too ! ### Objectives The objectives of this project are first and foremost educational, I wanted to build my own Spanish AI assistant to help me with my Spanish learning journey. To make that happen I set for myself the next set of goals : - Make a local and private AI assistant, no need for external APIs. (Hna gha talaba) - Ensure that the system understands my english AND Spanish questions (like a real teacher would) with Multi-lingual retrieval. - Have a minimal layer of (XAI) explainable AI so that whenever the assistant gives me an answer it also gives me the source of the Spanish rule or rules it used. - Have a smooth UI and a relatively short respond time. ### Tech Stack - Python : As the main programming language - Ollama : To run our AI model - Llama 3.1 : As our large language model (LLM). - LangChain : To smoothly connect the LLM to our data. - ChromaDB : For the vector database. - HuggingFace Multilingual MiniLM - Chainlit : For our minimalist chat interface. ### How to use My favourite part of project documentation is to show how the system can be used. It is a local AI assistant so you need to install quite few things on your laptop to be able to use it (sorry). 1. First : install Ollama and pull the model, i used llama3.1 ```bash curl -fsSL ollama.com | sh ollama run llama3.1 ``` 2. Set up the environnement ```bash python -m venv .venv source .venv/bin/activate pip install -r requirements.txt ``` 3. Ingest the Spanish rules t …