DomainSmith

DomainSmith is a technical platform designed to transform static documents—such as manuals, datasets, and internal documentation—into interactive, domain-aware AI assistants. It solves the problem of "context-blind" AI by implementing a local Retrieval-Augmented Generation (RAG) pipeline, allowing users to query their own data privately using local large language models (LLMs). 1

Features

• Universal Uploads: Support for PDF, Word, and plain text files that are parsed, chunked, and stored for retrieval
• Embeddings & Search: Local Ollama embeddings with per-domain collections for fast, relevant retrieval
• Chat & RAG: Ask natural questions with responses that reference relevant source documents for explainability
• Private & Local: Runs with your own Ollama and MongoDB—you keep control over data and models.

Tech Stack

Component	Technology	Role
Frontend	React + Vite	Single Page Application (SPA) for domain management and chat interface
Backend	Express.js	REST API handling authentication, file uploads, and RAG orchestration
Database	MongoDB	Stores user metadata, domain configurations, and vector embeddings
AI Engine	Ollama	Local service for generating embeddings and LLM chat completions

Prerequisites

• Node.js v18.0.0+
• MongoDB v6.0+
• Ollama (Latest)
• npm v9.0.0+

AI Model Setup

After installing Ollama, pull the required models:

ollama pull mistral
ollama pull nomic-embed-text

Installation

1. Clone the Repository

git clone https://github.com/AkiTheMemeGod/DomainSmith.git
cd DomainSmith

2. Backend Setup

cd backend
npm install

3. Frontend Setup

cd ../frontend
npm install

Configuration

Backend Configuration

Create a .env file in the backend/ directory:

PORT=5000
MONGO_URI=mongodb://localhost:27017/domainsmith
JWT_SECRET=your_super_secret_key
JWT_EXPIRES_IN=7d
OLLAMA_URL=http://127.0.0.1:11434
OLLAMA_MODEL=mistral
EMBEDDING_MODEL=nomic-embed-text
UPLOAD_DIR=./uploads

Frontend Configuration

Create a .env file in the frontend/ directory:

VITE_API_URL=http://localhost:5000/api

Running the Application

Start Backend

cd backend
npm run dev

Start Frontend

cd frontend
npm run dev

The frontend will be available at http://localhost:5173 and the backend API at http://localhost:5000.

Architecture

DomainSmith operates on the concept of a "Domain," which acts as a siloed knowledge base. The system follows a classic client-server architecture with a specialized AI integration layer.

RAG Pipeline

The core value of DomainSmith is its ability to bridge "Natural Language Space" and "Code Entity Space" through vectorization:

1. Document Ingestion: Users upload PDF, Word, or text files
2. Parsing: Documents are parsed using pdfParser or mammoth for Word documents
3. Chunking: Text is split into chunks using a sliding window chunker
4. Embedding: Chunks are converted to vectors using nomic-embed-text via Ollama
5. Storage: Vectors are stored in MongoDB for retrieval
6. Retrieval: When a user asks a question, the system retrieves relevant document chunks
7. Generation: Context is provided to a Mistral model to generate an informed response

Usage

1. Create an Account: Register a new account on the landing page
2. Create a Domain: Navigate to the dashboard and create a new domain
3. Upload Documents: Upload PDF, Word, or text files to your domain
4. Build Domain: Trigger the build process to parse, chunk, and embed your documents
5. Chat: Once the domain status is "ready", start chatting with your domain-aware AI assistant

Project Structure

DomainSmith/
├── backend/
│   ├── src/
│   │   ├── config/
│   │   ├── models/
│   │   ├── routes/
│   │   ├── services/
│   │   └── utils/
│   └── package.json
└── frontend/
    ├── src/
    │   ├── api/
    │   ├── pages/
    │   └── App.jsx
    └── package.json

Notes

This README is based on the DomainSmith Overview and Getting Started wiki pages. For more detailed technical documentation, refer to the wiki pages in the repository. The project uses local Ollama models (mistral for chat and nomic-embed-text for embeddings) which must be pulled separately before running the application. 6