🧠 AutoDocThinker: Agentic RAG System with Intelligent Search Engine

The Agentic RAG System is a full-stack, AI-powered multi-agent document intelligence platform that extracts insights from PDFs, DOCX, TXT files, and web URLs through natural language queries. Built with Python, Flask, LangChain, ChromaDB, and HuggingFace embeddings, it orchestrates modular agents for document ingestion, chunking, vector storage, context retrieval, LLM reasoning, and Wikipedia fallback.

The system achieves ~95% context retrieval coverage, generates answers with ~70% accuracy and F1 ~68%, and triggers fallback in only 5% of queries, demonstrating robust reliability. Intelligent agent routing improves query efficiency by ~40%, and the system scales to handle 50+ documents simultaneously. With a responsive HTML/CSS/Bootstrap UI, secure file handling, modular backend, logging, and Docker deployment, the platform delivers measurable business impact through fast, accurate, and scalable document intelligence.

demo.mp4

---

🚀 Live Demo

🖥️ Try it now: AutoDocThinker: Agentic RAG System with Intelligent Search Engine

---

⚙️ Features & Functionalities

#	Module	Technology Stack	Your Implementation Details
1	LLM Processing	Groq + LLaMA-3-70B	Configured with optimal temperature (0.2) and token limits
2	Document Parsing	PyMuPDF + python-docx	Handled PDF, DOCX, TXT with metadata preservation
3	Text Chunking	RecursiveCharacterTextSplitter	500-character chunks with 20% overlap for context
4	Vector Embeddings	all-MiniLM-L6-v2	Efficient 384-dimensional embeddings
5	Vector Database	ChromaDB	Local persistent storage with cosine similarity
6	Agent Workflow	LangGraph	7 specialized nodes with conditional routing
7	Planner Agent	LangGraph Planner Node	Generates execution plans
8	Executor Agent	LangGraph Node	Orchestrates tool calls
9	Web Fallback	Wikipedia API	Auto-triggered when document confidence < threshold
10	Memory System	deque(maxlen=3)	Maintained conversation history buffer
11	User Interface	HTML, CSS, Bootstrap, JS	Interactive web app with file, URL, Text upload
12	Containerization	Docker	Portable deployment
13	CI/CD Pipeline	GitHub Actions	Automated linting/testing

---

🧱 Project Structure

AutoDocThinker/
├── .github/
│ └── workflows/
│     └── main.yml
│  
├── agents/
│ ├── init.py
│ ├── document_processor.py
│ └── orchestration.py
│  
├── data/
│ └── sample.pdf
│  
├── notebooks/
│ └── experiment.ipynb
│  
├── static/
│ ├── css/
│ │ └── style.css
│ └── js/
│   └── script.js
│  
├── templates/
│ └── index.html
│  
├── tests/
│ └── test_app.py
│  
├── uploads/
│  
├── vector_db/
│ └── chroma_collection/
│   └── chroma.sqlite3
│
├── app.log
├── app.py
├── demo.mp4
├── demo.png
├── Dockerfile
├── LICENSE
├── render.yaml
├── README.md
├── requirements.txt
└── setup.py

---

🧱 System Architecture

%% Agentic RAG System Architecture - Colorful Version
graph TD
    A[User Interface]:::ui -->|Upload/Input| B[Flask Web Server]:::server
    B --> C[Tool Router Agent]:::router
    C -->|File| D[Document Processor]:::processor
    C -->|URL| E[Web Scraper]:::scraper
    C -->|Text| F[Text Preprocessor]:::preprocessor
    
    D --> G[PDF/DOCX/TXT Parser]:::parser
    E --> H[URL Content Extractor]:::extractor
    F --> I[Text Chunker]:::chunker
    
    G --> J[Chunking & Embedding]:::embedding
    H --> J
    I --> J
    
    J --> K[Vector Database]:::database
    
    B -->|Query| L[Planner Agent]:::planner
    L -->|Has Documents| M[Retriever Agent]:::retriever
    L -->|No Documents| N[Fallback Agent]:::fallback
    
    M --> K
    K --> O[LLM Answer Agent]:::llm
    N --> P[Wikipedia API]:::api
    P --> O
    
    O --> Q[Response Formatter]:::formatter
    Q --> B
    B --> A

    classDef ui fill:#4e79a7,color:white,stroke:#333;
    classDef server fill:#f28e2b,color:white,stroke:#333;
    classDef router fill:#e15759,color:white,stroke:#333;
    classDef processor fill:#76b7b2,color:white,stroke:#333;
    classDef scraper fill:#59a14f,color:white,stroke:#333;
    classDef preprocessor fill:#edc948,color:#333,stroke:#333;
    classDef parser fill:#b07aa1,color:white,stroke:#333;
    classDef extractor fill:#ff9da7,color:#333,stroke:#333;
    classDef chunker fill:#9c755f,color:white,stroke:#333;
    classDef embedding fill:#bab0ac,color:#333,stroke:#333;
    classDef database fill:#8cd17d,color:#333,stroke:#333;
    classDef planner fill:#499894,color:white,stroke:#333;
    classDef retriever fill:#86bcb6,color:#333,stroke:#333;
    classDef fallback fill:#f1ce63,color:#333,stroke:#333;
    classDef llm fill:#d37295,color:white,stroke:#333;
    classDef api fill:#a0d6e5,color:#333,stroke:#333;
    classDef formatter fill:#b3b3b3,color:#333,stroke:#333;

Loading

---

🌍 Real-World Applications

1. Corporate HR Automation
2. Legal Document Review
3. Academic Research
4. Customer Support
5. Healthcare Compliance
6. Financial Analysis
7. Media Monitoring
8. Education
9. Technical Documentation
10. Government Transparency

---

📥 Installation

# 1. Clone the repository
git clone https://github.com/Md-Emon-Hasan/AutoDocThinker.git
cd AutoDocThinker

# 2. Install dependencies
pip install -r requirements.txt

Or with Docker:

# Build Docker Image
docker build -t auto-doc-thinker .

# Run the container
docker run -p 8501:8501 auto-doc-thinker

---

🔁 GitHub Actions CI/CD

.github/workflows/main.yml

name: CI

on:
  push:
    branches: [ master ]
  pull_request:
    branches: [ master ]

jobs:
  build-and-test:
    runs-on: ubuntu-latest

    steps:
      - uses: actions/checkout@v3
      - name: Set up Python
        uses: actions/setup-python@v4
        with:
          python-version: "3.11"
      - name: Install dependencies
        run: |
          pip install -r requirements.txt
      - name: Lint with flake8
        run: |
          pip install flake8
          flake8 .

---

📝 Future Enhancements

• ✅ Multilingual document ingestion
• ✅ Audio document ingestion + whisper
• ⏳ Long-term memory + history viewer
• ⏳ MongoDB/FAISS alternative for Chroma
• ✅ More tools (WolframAlpha, SerpAPI)
• ⏳ Model selection dropdown (Gemini, LLaMA, GPT-4)

---

👨‍💻 Author

Md Emon Hasan 📧 Email: email 🔗 LinkedIn: md-emon-hasan 🔗 GitHub: Md-Emon-Hasan 🔗 Facebook: mdemon.hasan2001/ 🔗 WhatsApp: 8801834363533

---