🛍️ Multimodal Product Recommendation System

An AI-powered product recommendation system that combines:

• Gemini Vision
• ChromaDB Vector Search
• Retrieval-Augmented Generation (RAG)
• Streamlit UI

Users can upload a product image and receive similar product recommendations using multimodal AI.

---

Features

✅ Product image upload

✅ Gemini Vision image understanding

✅ Semantic vector search using ChromaDB

✅ Retrieval-Augmented Generation (RAG)

✅ AI-generated product recommendations

✅ Streamlit web interface

✅ Docker support

---

Architecture

User Uploads Image ↓ Gemini Vision ↓ Image Description ↓ Embedding Generation ↓ ChromaDB Retrieval ↓ Top Similar Products ↓ Gemini Recommendation Generation ↓ Streamlit UI

---

Tech Stack

• Python
• Gemini API
• ChromaDB
• Streamlit
• Docker
• Vector Embeddings

---

Project Structure

.
├── app
│   ├── vision.py
│   ├── retrieval.py
│   ├── generator.py
│   ├── multimodal_agent.py
│
├── data
│
├── tests
│
├── vector_db
│
├── streamlit_app.py
│
├── requirements.txt
│
└── README.md

Installation

Clone Repository

git clone <repo-url>
cd <repo-name>

Create Virtual Environment

python -m venv venv
source venv/bin/activate

Install Dependencies

pip install -r requirements.txt

Environment Variables

Create a .env file:

GOOGLE_API_KEY=your_api_key_here

---

Running the Application

streamlit run streamlit_app.py

Open:

http://localhost:8501

---

Example Workflow

1. Upload a product image.
2. Gemini Vision analyzes the image.
3. Product description is generated.
4. ChromaDB retrieves similar products.
5. Gemini generates recommendations.
6. Results are displayed in Streamlit.

---

Future Improvements

• AWS Deployment
• Product Re-ranking
• Hybrid Search
• Agentic Workflow
• Advanced Evaluation Metrics

---

Author

Akash Kanwar