Knowledge Augmented Gen - 📄 A Private Document Summarizer using RAG, LangChain & IBM Watsonx

🚀 Objective

This project solves the problem of summarizing and querying private documents securely and efficiently using Retrieval-Augmented Generation (RAG) architecture.
Instead of uploading sensitive documents to public LLM services, this solution:

• Stores & indexes documents locally.
• Converts them into embeddings for efficient search.
• Retrieves only relevant information chunks.
• Sends only small, relevant snippets to the cloud-based IBM Watsonx.ai LLM for question answering.

Ideal use-case: Onboarding at a company with a large volume of private policies, guidelines, or project documents you need to quickly understand, but can’t risk uploading publicly.

---

🛠️ Technologies Used

Technology	Purpose
LangChain	Document loading, splitting, RAG pipeline
Chroma DB	Local vector storage for embeddings
Hugging Face Embeddings	Convert documents to numerical vectors
IBM Watsonx.ai LLM	Language model for summarization and QA
dotenv	Securely load API keys and configs
Python 3.8+	Language

---

📂 Project Structure

rag_summarizer/
├── data/                         # Private documents
│   └── companyPolicies.txt
├── config/                       # Environment configs
│   └── .env
├── embeddings/                   # Embedding-related functions
│   └── embedder.py
├── retriever/                    # Retriever setup
│   └── retriever.py
├── llm/                          # Watsonx LLM setup
│   └── watsonx_llm.py
├── chain/                        # Building the RAG conversation chain
│   └── qa_chain.py
├── utils/                        # Helpers (optional)
├── main.py                       # Entry point
├── README.md                     # Project documentation
└── requirements.txt              # Dependencies

---

🔑 Environment Setup

1. Clone the Repository

git clone https://github.com/yourusername/Knowledge-Augmented-Gen.git
cd Knowledge-Augmented-Gen

2. Install Dependencies

pip install -r requirements.txt

3. Setup .env File

Inside /config/.env:

IBM_URL=your_ibm_url
IBM_API_KEY=your_ibm_api_key
IBM_PROJECT_ID=your_project_id

📝 Ensure your IBM Watsonx credentials are secured here.

---

📝 How It Works

1. Indexing Phase:

• Load documents: .txt files (company policies, guidelines, manuals, etc.)
• Split into chunks: Using CharacterTextSplitter for manageable processing.
• Convert to embeddings: Using Hugging Face models.
• Store locally: In Chroma DB for fast vector-based retrieval.

---

2. Retrieval & QA Phase:

• Query initiated by user.
• Relevant text chunks retrieved locally.
• Only relevant snippets are sent to IBM Watsonx.ai LLM.
• Returns accurate, contextual, and private responses.

---

3. Conversation Memory:

• Retains multi-turn conversation history, allowing context-aware follow-up questions.

---

💻 Usage

Run the project:

python main.py

You'll enter an interactive prompt:

Welcome! Ask your questions (type 'exit' to quit):
Question: What is the mobile policy?
Answer: [LLM-generated answer]

Question: What is the aim of it?
Answer: [Context-aware response]

Type exit to quit.

---

🧩 Customization Ideas

• Replace IBM Watsonx with Local LLMs (Mistral, Llama2, GPT4All).
• Expand for PDF/CSV documents.
• Add Flask or Streamlit UI.
• Integrate logging and analytics for usage tracking.

---

📌 Why This Project Matters

• Keeps private documents secure & local.
• Minimizes data exposure.
• Leverages state-of-the-art RAG architecture.
• Easily extensible and production-ready.

---

📜 License

MIT License – feel free to fork, modify, and build upon!

---

✨ Contributors

• [Md Yeasin Arafath] (Developer, Architect)