AIJStack

AI stack built on Node.js and powered by RAG Retrieval-Augmented Generation

Stack architecture

The below table is describing the different layer abstractions

Layer	Node.js Tool/Lib	Notes
Document loader	fs	-
Document Parser	langchain	File format supported txt
RAG Framework	langchain	Chain together: Retrieval + Generation
RAG Embeddings	Xenova/all-MiniLM-L6-v2	Fast 384-dim sentence embeddings
LLM	EleutherAI/gpt-neo-125M	Lightweight Model

Initally based on chatGPT discussion, 27-06-2025

Example project instance

The current instance is running a generative chat supplying basic response on a specific theme

Prerequirements

• Python 3.9, later version are not yet covered by torch prebuilt
• Nodejs version >= 18.0 (not tested)
• English documentation is only supported by LLM

Installation steps

Perform global installation

Use the installation script which is performing the following sub scripts,

• _js-env_install script dedicated to node environment
• _py-env_install script dedicated to python environment

Note,

py-env_install script is creating a .venv virtual environment and imports the following packages (fastapi, uvicorn, torch, transformers, accelerate, pydantic)

How to run it?

1. Dispose your documentation into the /doc folder. Note: Currently only 'txt' files are supported
2. Run both nodejs and model script instances with quick-start or npm run start-dev:all
3. Send a POST HTTP JSON message to http://localhost:3000/generate i.e : {"query" : "What is (your scope) ?"}
4. Expecting for a HTTP '201' JSON message response (json anwser response : Context... > Question... > Answer...)