https://github.com/udit-rawat/contextcraft
A RAG project that explores RAG technology and uses ollama to demonstrate its functionality on local system.
https://github.com/udit-rawat/contextcraft
llama3 ollama rag sentence-transformers streamlit
Last synced: about 2 months ago
JSON representation
A RAG project that explores RAG technology and uses ollama to demonstrate its functionality on local system.
- Host: GitHub
- URL: https://github.com/udit-rawat/contextcraft
- Owner: udit-rawat
- License: unlicense
- Created: 2024-11-25T17:25:00.000Z (over 1 year ago)
- Default Branch: main
- Last Pushed: 2024-11-25T17:59:05.000Z (over 1 year ago)
- Last Synced: 2025-10-30T01:45:03.462Z (8 months ago)
- Topics: llama3, ollama, rag, sentence-transformers, streamlit
- Language: Jupyter Notebook
- Homepage:
- Size: 231 KB
- Stars: 1
- Watchers: 1
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# ContextCraft: A Retrieval-Augmented Generation Framework
## 📚 Project Overview
**ContextCraft** is a Retrieval-Augmented Generation (RAG) framework designed for intelligent query answering and document analysis. It integrates various components like PDF parsing, vector generation, FAISS indexing, and response generation to create an efficient pipeline for extracting insights from large documents.
### 🛠️ Key Features
- **PDF Parsing and Chunking**: Breaks down documents into manageable chunks for processing.
- **Vector Store Integration**: Uses FAISS for efficient similarity search and indexing.
- **State-of-the-Art Models**: Leverages `all-mpnet-base-v2` for vector generation and `llama3.2:latest` for generating intelligent responses.
- **Streamlit UI**: Provides an interactive web application for user-friendly query answering.
---
## ⚙️ Workflow
The ContextCraft pipeline follows the stages depicted in the diagram:
1. **PDF Parsing and Chunk Creation**:
- Input PDF files are parsed and split into smaller chunks for efficient processing.
2. **Vector Generation**:
- Uses the `all-mpnet-base-v2` model to generate embeddings for each document chunk.
3. **Vector Store**:
- Embeddings are stored in a FAISS index for fast similarity searches.
4. **Query Engine**:
- Facilitates retrieval of the most relevant chunks using k-retrieval.
5. **Retriever and Content Engine**:
- Aggregates retrieved chunks and prepares context for generating responses.
6. **Response Generation**:
- Powered by `llama3.2:latest`, the system generates natural language answers based on the retrieved context.
7. **Streamlit Application**:
- A user-friendly web interface built with Streamlit (`app.py`) for real-time interaction.
---
## 🚀 Installation
### Prerequisites
- Python >= 3.11
- Virtual environment (venv)
- Jupyter Notebook or JupyterLab
- Streamlit for running the web app
### Steps
1. Clone the repository:
```bash
git clone https://github.com/udit-rawat/ContextCraft.git
cd ContextCraft
```
2. Create and activate a virtual environment:
- **On macOS/Linux**:
```bash
python3 -m venv venv
source venv/bin/activate
```
- **On Windows**:
```bash
python -m venv venv
.\venv\Scripts\activate
```
3. Install dependencies from `requirements.txt`:
```bash
pip install -r requirements.txt
```
4. Unzip your PDF file(s):
- Place your PDF files in the appropriate folder within the project directory.
5. Run the Jupyter Notebooks:
- Launch Jupyter Notebook from the command line:
```bash
jupyter notebook
```
- Open and run the following notebooks sequentially in the browser:
1. **`parser.ipynb`**: This notebook will parse your PDF file(s) and create chunks.
2. **`query_engine.ipynb`**: This notebook sets up the query engine for retrieving relevant document chunks.
3. **`vector_generator.ipynb`**: Generates vector embeddings for the chunks using the `all-mpnet-base-v2` model.
4. **`vector_store.ipynb`**: Stores the generated vectors in a FAISS index for fast retrieval.
---
## 🔧 Configuration
### Modify Paths
- Ensure that your PDF files are unzipped and placed in the correct directory as specified in the notebooks.
- Adjust FAISS index parameters if working with large datasets.
---
## 🌟 Key Technologies Used
- **FAISS**: Efficient similarity search and clustering of dense vectors.
- **Sentence Transformers**: `all-mpnet-base-v2` for embedding generation.
- **Llama Model**: `llama3.2:latest` for response generation.
- **Streamlit**: Fast and customizable web-based interface.
---
## 📂 Directory Structure
```
ContextCraft/
├── LICENSE # License file for the project
├── README.md # Documentation for the project
├── __pycache__/ # Compiled Python files
├── app.py # Main Streamlit application
├── contexcraft/ # Core functionality and code (venv)
├── embeddings/ # Folder for storing generated embeddings
├── notebooks/ # Folder for Jupyter Notebooks
│ ├── parser.ipynb # Notebook for PDF parsing and chunk creation
│ ├── query_engine.ipynb # Notebook for setting up query engine
│ ├── vector_generator.ipynb # Notebook for generating vectors
│ └── vector_store.ipynb # Notebook for storing vectors in FAISS
├── overview.png # Architecture diagram for ContextCraft
├── parsed_content/ # Folder for parsed content from PDFs
├── pdfs.zip # Compressed PDF files to be processed
├── pdfs/ # Folder for storing PDFs to be parsed
├── requirements.txt # List of dependencies for the project
├── retriever.py # Script for the retriever function
└── vector_store/ # Folder for storing FAISS index and vector data
```
---
## 📊 Example Use Case
1. Upload and unzip your PDF document in the `data/` directory.
2. Open and run the following notebooks in order:
- **`parser.ipynb`**: Parse the PDF and generate document chunks.
- **`query_engine.ipynb`**: Set up the query engine.
- **`vector_generator.ipynb`**: Generate vector embeddings for the document chunks.
- **`vector_store.ipynb`**: Store the embeddings in a FAISS index.
3. The system will process the PDF, generate embeddings, store them in FAISS, and allow for efficient query answering.
---
## 🤝 Contributing
1. Fork the repository.
2. Create a new branch:
```bash
git checkout -b feature-name
```
3. Commit your changes and create a pull request.
4. Refer to the whitepaper pdf provided for more ways to experiment with this project.
---
## 📜 License
This project is licensed under the MIT License.
---