RAG Knowledge Assistant

Overview

This project implements a Retrieval-Augmented Generation (RAG) system to create a self-hosted knowledge assistant. Users can upload their own documents and query them using natural language. The system is designed to run entirely locally, leveraging a locally hosted LLM (DeepSeek) and a Postgres database — no external APIs or cloud services required.

---

Features

• Upload Knowledge: Users can upload custom knowledge files (text, JSON, etc.).
• Natural Language Queries: Users can prompt the model with questions related to their uploaded knowledge.
• AI-Powered Responses: Powered by a local DeepSeek model, the system retrieves and generates contextually relevant answers without internet dependency.
• Fully Local Setup: No external API keys or model hosting required.
• Responsive UI: Built using Angular 19 with Material UI for a modern user experience.

---

Tech Stack

• Frontend: Angular 19, Material UI
• Backend: Node.js, Express, TypeScript (Domain-Driven Design)
• AI Service: FastAPI (Python) for generating embeddings and interacting with a locally hosted DeepSeek LLM
• Database: PostgreSQL for storing knowledge files and metadata
• Storage: File system (future plans to support S3-compatible storage)

---

Local Setup and Installation

1. Clone the repository

git clone https://github.com/aroo530/Knowledge-hub.git
cd Knowledge-hub

2. Install dependencies

• Angular frontend:

  cd angular-app
  npm install

• Node.js backend:

  cd ../backend
  npm install

• Python FastAPI service:

  pip install -r requirements.txt

🧠 Make sure the DeepSeek model is downloaded and placed in pythonEmbeddingsGen/models. and add the model name in the Python code

3. Start the services

• Frontend (Angular):

  cd angular-app
  ng serve

• Backend (Node.js/Express):

  cd ../backend
  npm start

• AI Service (FastAPI):

  cd pythonEmbeddingsGen
  uvicorn main:app --reload

4. Access the app

• Frontend: http://localhost:4200
• Backend: http://localhost:3000
• AI Service: http://localhost:8000

---

Usage

1. Upload your knowledge base via the "Upload Knowledge" interface.
2. Ask questions in natural language through the chat interface.
3. The system uses local embeddings and a local LLM to retrieve relevant content and respond intelligently.

---

Why Local?

• Privacy-first: No data is sent to third-party services.
• Offline capability: Works in completely air-gapped environments.
• Cost-effective: No need for API keys, model hosting, or cloud compute.

---

Future Improvements

• Cloud file storage support (e.g. S3, MinIO)
• Multi-user management and auth
• Improved chunking and context retrieval
• Model upgrade to support longer contexts and multi-modal input

---