AI Vision RAG Explorer

An interactive Visual Retrieval-Augmented Generation (RAG) system that combines:

🔎 Cohere Embed-4 for multimodal embeddings

⚡ Google Gemini 2.5 Flash for visual question answering

Built with Streamlit, this app allows you to upload images and PDFs, then query them with natural language to extract insights from charts, diagrams, and document pages.

✨ Features

📂 Upload PDFs & Images

• PDFs are automatically converted into page images.
• Images are auto-enhanced for better embedding + Q&A.

🔎 Multimodal Retrieval

• Uses Cohere Embed-4 to compute embeddings for each image/page.
• Finds the most semantically relevant page/image for a given query.

🤖 AI-Powered Answers

• Google Gemini 2.5 Flash analyzes the retrieved visual content.
• Generates clear, context-aware answers to your question.

🎨 Modern UI/UX

• Gradient background, styled buttons, answer bubbles, image cards.
• Dual-column layout → relevant image on the left, Gemini answer on the right.

🧠 Session Management

• Stores uploaded files and embeddings in Streamlit session state.
• Enables multiple queries without re-uploading documents.

⚙️ Requirements

• Python 3.9+
• Cohere API key
• Google Gemini API key

🚀 Installation & Setup

1. Create Virtual Environment

python -m venv venv
.\venv\Scripts\activate

2. Install Dependencies

pip install --upgrade pip
pip install -r requirements.txt
pip install streamlit cohere google-genai python-dotenv PyMuPDF pillow numpy

3. Add API Keys

Create a .env file in the root folder:

COHERE_API_KEY=your_cohere_api_key
GEMINI_API_KEY=your_gemini_api_key

4. Run Streamlit App

streamlit run app.py

🔍 How It Works

1. 📥 Upload PDFs or Images

• PDFs → Each page is rendered to an image using PyMuPDF.
• Images → Automatically enhanced and resized for better processing.

2. 🧮 Embedding with Cohere Embed-4

• Generates dense multimodal embeddings for every image or PDF page.

3. ❓ Ask a Natural Language Question

• Your query is embedded using search_query mode.
• Cosine similarity retrieves the most relevant image/page.

4. ⚡ Answer Generation with Gemini

• The retrieved image + your question are analyzed using Google Gemini.
• Produces a context-aware AI-generated answer.

5. 🖼️ Results

• Left panel → Relevant image/page
• Right panel → Gemini’s generated answer

📌 Example Use Cases

• 📊 Extract insights from financial charts
• 📑 Understand tables and diagrams in PDFs
• 🔎 Ask targeted questions about multi-page documents
• 🗂️ Perform visual knowledge retrieval from mixed content (images + PDFs)