📄 GPT-4ALL Document Query Chatbot

This project enables users to query a wide variety of documents using an advanced chatbot powered by open-source LLMs like GPT-4ALL and Llama. Leveraging embeddings, vector databases, and data loaders, this system efficiently handles document parsing, storage, and retrieval.

📑 Table of Contents

• 📌 Project Overview
• 🚀 Features
• 📂 Project Structure
• 🛠 Installation
• 🚀 Usage
• 🔗 General Links & Resources
• ⚙️ Configuration
• 🗂️ Supported Document Formats
• 📈 Limitations & Next Steps
• 📄 License
• 📞 Support

---

📌 Project Overview

In today’s data-intensive environments, there’s a growing need to convert unstructured data into actionable insights. This chatbot bridges that gap by allowing users to interactively query documents, with support for multiple formats including PDF, Word, PowerPoint, Markdown, and more.

Built with langchain and chromadb, this solution processes documents by:

• Converting them into text chunks.
• Embedding these chunks as vectors.
• Storing them for easy retrieval, powered by a selected LLM model.

---

🚀 Features

• Multi-format Document Support: Accepts documents in .pdf, .docx, .pptx, .txt, and other formats.
• Embeddings with Langchain: Uses HuggingFaceBgeEmbeddings for text chunk embeddings.
• Vector Storage with ChromaDB: Stores text embeddings as vectors for efficient retrieval.
• Choice of LLMs: Supports GPT-4ALL and Llama models for answering queries.
• Customizable Environment: Easily configure model and embedding options via .env.

---

📂 Project Structure

• requirements.txt: Lists necessary Python packages.
• .env: Contains environment variables for model and database settings.
• constants.py: Holds constants for Chroma database configuration.
• ingest.py: Processes and stores documents as vectors for future querying.
• privateGPT.py: Main chatbot script for querying stored documents.

---

🛠 Installation

1. Clone the repository:

   git clone https://github.com/MoAshour93/Construction_Private_GPT.git
   cd Construction_Private_GPT

2. Install dependencies:

   pip install -r requirements.txt

3. Set up environment variables:

   • Create a .env file in the root directory, using the provided template:

     PERSIST_DIRECTORY=db
     MODEL_TYPE=GPT4All
     MODEL_PATH=models/ggml-gpt4all-j-v1.3-groovy.bin
     EMBEDDINGS_MODEL_NAME=all-MiniLM-L6-v2
     MODEL_N_CTX=1000
     MODEL_N_BATCH=8
     TARGET_SOURCE_CHUNKS=4

---

🚀 Usage

1. Ingest Documents

Use ingest.py to process and store document embeddings:

python ingest.py

2. Run the Chatbot

Start querying documents using privateGPT.py:

python privateGPT.py

• Enter your query at the prompt.
• Type exit to end the session.

🔧 Customizable Options

• Use --hide-source or -S to hide source documents used in responses.
• Use --mute-stream or -M to disable streaming output from the LLM.

---

🔗 General Links & Resources

• Our Website: www.apcmasterypath.co.uk
• APC Mastery Path Blogposts: APC Blogposts
• LinkedIn Pages: Personal | APC Mastery Path

---

⚙️ Configuration

• Constants: The constants.py file includes important settings for the ChromaDB database.
• Environment Variables: Set customizable parameters in .env, including model path and embedding model name.

---

🗂️ Supported Document Formats

Format	Loader
PDF	PyPDFLoader
Word Documents	UnstructuredWordDocumentLoader
PowerPoint	UnstructuredPowerPointLoader
Markdown	UnstructuredMarkdownLoader
CSV	CSVLoader
Text	TextLoader

---

📈 Limitations & Next Steps

This initial implementation is a command-line-based chatbot, but it can be extended:

1. GUI Integration: Integrate with Streamlit or Chainlit for a graphical user interface.
2. Multi-agent Architecture: Develop task-specific agents for more complex queries.
3. Broader LLM Support: Experiment with other open-source models from Hugging Face.

---

📄 License

This project is licensed under the Apache 2.0 License.

---

📞 Support

For any questions, feel free to contact Mohamed Ashour.