🧠 war-rag: Retrieval-Augmented Generation with The Art of War

This project demonstrates an end-to-end RAG (Retrieval-Augmented Generation) pipeline using LangChain, FAISS, and a local LLM (orca-mini via Ollama). The source content is Sun Tzu's The Art of War (public domain via Project Gutenberg).

---

📐 Project Goals

• Understand each stage of RAG: Ingest → Embed → Retrieve → Augment → Generate
• Use local models for embeddings and generation (no OpenAI needed)
• Build everything step by step inside a .ipynb notebook
• Prioritize clarity over complexity — one concept per cell

---

🧱 Stack

Component	Technology
Text Source	Project Gutenberg
Chunking	RecursiveCharacterTextSplitter
Embeddings	nomic-embed-text via Ollama
Vector DB	FAISS
LLM	orca-mini via Ollama
Frameworks	LangChain, LangGraph (optional)
Interface	Jupyter Notebook
Environment	Conda + .env configuration

---

🔁 RAG Pipeline Overview

🧠 Conceptual Flow

flowchart TD
    A[User Query] --> B[Embed Query]
    B --> C[FAISS Similarity Search]
    C --> D[Retrieve Top-K Chunks]
    D --> E[Build Prompt]
    E --> F[Generate Answer via LLM]
    F --> G[Final Answer]

Loading

🧪 Execution Phases

graph TB
    subgraph Preprocessing
        A1[Download & Clean Text]
        A2[Chunk into Documents]
        A3[Embed Chunks]
        A4[Store in FAISS]
    end

    subgraph Runtime
        B1[User Query]
        B2[Embed Query]
        B3[Search FAISS]
        B4[Retrieve Top-K]
        B5[Build Prompt]
        B6[Generate Answer]
    end

    A1 --> A2 --> A3 --> A4
    B1 --> B2 --> B3 --> B4 --> B5 --> B6

Loading

---

📓 Step-by-Step Summary

Step	Purpose
1	Download and clean The Art of War text
2	Chunk the document using RecursiveCharacterTextSplitter
3	Inspect chunk overlap
4	Embed each chunk via nomic-embed-text
5	Query FAISS with user input
6	Build a prompt from retrieved context
7	Pass prompt to orca-mini and generate grounded answer

---

🧪 Example Query

Question: "What does Sun Tzu say about deception?"
Generated Answer: According to Sun Tzu in The Art of War, deception plays a crucial role in warfare... (retrieved from vector store and grounded)

---

🔧 Local Setup

# Clone repo (if extracted elsewhere)
git clone <your-repo>
cd war-rag

# Create conda environment
conda create -n war-rag python=3.10
conda activate war-rag

# Install dependencies
pip install -r requirements.txt

# Pull models (if needed)
ollama pull nomic-embed-text
ollama pull orca-mini

# Run notebook
jupyter lab

.env file example:

EMBEDDING_MODEL=nomic-embed-text

---

🚀 Future Enhancements

• Add LangGraph orchestration of RAG flow
• Add streaming token output via FastAPI/CLI
• Build structured answer formatting or summarization layer
• Turn into a deployable chatbot or research assistant

---

📜 License

Content from The Art of War is in the public domain. This project is intended for learning and experimentation.