💎 InsightDoc AI | Document & Visual Intelligence Suite

InsightDoc AI is a general-purpose Document & Visual Intelligence Suite built on a Retrieval-Augmented Generation (RAG) architecture.
It delivers precise, context-faithful answers by combining semantic retrieval, entity-aware filtering, and visual (OCR-based) understanding.

---

🎥 Demo Video

InsightDoc.mp4

---

🧠 System Overview

• Architecture: Retrieval-Augmented Generation (RAG)
• LLM: Meta LLaMA-3 (via Groq)
• Vector Store: ChromaDB (persistent, local)
• Embeddings: HuggingFace all-MiniLM-L6-v2
• Evaluation: RAGAS (precision, recall, faithfulness, relevancy)

---

✨ Key Features

• 📄 Document Intelligence
  Understands and answers questions grounded strictly in retrieved document context.

• 👁️ Visual Intelligence
  OCR-enabled reasoning over visual inputs for text-based understanding.

• 🧠 Entity-Aware Context Isolation
  Automatically detects the target entity in a query and restricts retrieval to relevant context only.

• 🔍 Hybrid Retrieval Strategy
  Combines semantic similarity with metadata validation to reduce cross-document leakage.

• 🧾 Context-Only Answering
  Generates answers strictly from retrieved context with clear fallbacks when information is missing.

• 📊 Evaluation-Ready RAG Pipeline
  Integrated RAGAS evaluation to measure real retrieval and generation quality.

• 🔐 Local & Secure Indexing
  All embeddings and vectors are stored locally using ChromaDB.

---

⚙️ Design Enhancements

• Robust entity extraction with filename fallback
• Entity normalization for reliable filtering
• Third-person answer normalization
• Hybrid metadata + content validation
• Prevents artificially inflated RAG evaluation scores

---

🛠 Tech Stack

Layer	Tools Used
Backend	FastAPI
Frontend	Streamlit
LLM	Groq (LLaMA-3)
Embeddings	HuggingFace MiniLM
Vector DB	ChromaDB
OCR	Tesseract
Evaluation	RAGAS
Language	Python

---

🚀 How to Run Locally

1️⃣ Clone the repository

git clone https://github.com/Its-Itachi/InsightDoc-AI.git
cd InsightDoc-AI

2️⃣ Create a virtual environment

python -m venv venv

3️⃣ Activate the virtual environment

Windows (PowerShell):

venv\Scripts\activate

macOS / Linux:

source venv/bin/activate

4️⃣ Install dependencies

pip install -r requirements.txt

---

🔑 Environment Configuration

Create a .env file in the project root:

GROQ_API_KEY=your_groq_api_key_here
TOKENIZERS_PARALLELISM=false

---

▶ Running the Application

Start Backend (FastAPI)

uvicorn main:app --reload --port 8000

API Docs:

http://localhost:8000/docs

---

Start Frontend (Streamlit)

streamlit run streamlit_app.py

Application UI:

http://localhost:8501

---

📊 RAG Evaluation (RAGAS)

InsightDoc AI includes an evaluation pipeline using RAGAS to measure:

• Context Precision
• Context Recall
• Faithfulness
• Answer Relevancy

Run evaluation:

python evaluation/ragas_eval.py

---

👤 Author

Jayesh Dethe

GitHub: https://github.com/Its-Itachi

---

📝 Notes

• Designed to reflect production-grade document intelligence
• Prevents cross-document and cross-entity hallucination

---

Happy coding! 🚀