👚 AI Fashion Visual Search & Intelligent Styling Assistant

An AI-powered fashion assistant that helps users discover visually similar fashion products, get intelligent outfit suggestions, and receive personalized recommendations — all within an elegant Streamlit app.

Think of it as your AI stylist: Upload a fashion product → Get similar items → Complete your look → Discover trending styles.

---

📷 Screenshots

Here are some snapshots of the Fashion Visual Search & Styling Assistant in action:

Visually Similar Products based on user image query	Suggestions to Complete Outfit	Products Based on Text Query	Personalized Recommendations Based on Your Style History

---

🌟 Key Features

🔍 1. Visual Similarity Search

Using CLIP embeddings + FAISS indexing, the system matches a user-uploaded product image with visually similar products from your fashion catalog (image_embeddings.pkl).

• Efficient vector similarity search via combined_vectors.npy
• Product details fetched via product_metadata_df.pkl
• Supports all major fashion categories (tops, pants, shoes, etc.)

---

🧠 2. Trend-Based Outfit Suggestions

The assistant uses an LLM (Gemma-3B) to generate outfit recommendations based on:

• The uploaded product's type, color, and pattern

• Trending clothing keywords dynamically extracted from fashion blogs and websites (trend_string.pkl)

• Language-enhanced responses like:

  “Pair this pastel top with high-waisted beige trousers and white sneakers for a clean summer look.”

---

👤 3. Personalized Styling using User History

By maintaining user_history.pkl, the system personalizes recommendations:

• Suggests items from similar clusters (e.g., casual, party, boho) using cluster_labels.pkl
• Recalls recently viewed/saved styles
• Avoids redundancy in recommendations
• Incorporates trending styles while aligning with the user’s historical preferences

---

💬 4. Text-Based Search (Coming Soon)

Users will be able to input natural language queries like:

"Show me casual outfits with floral tops under ₹1500" and get curated product results.

---

🛠 Project Structure

📁 Assets/
│   ├── cluster_labels.pkl             ← Cluster ID for each product
│   ├── combined_vectors.npy           ← Merged image-text embeddings
│   ├── faiss_index.index              ← FAISS index for fast search
│   ├── image_embeddings.pkl           ← CLIP-based image vectors
│   ├── text_embeddings.pkl            ← SentenceTransformer vectors
│   ├── product_metadata_df.pkl        ← All product details
│   ├── product_ids.pkl                ← ID mapping for inventory
│   ├── trend_string.pkl               ← Extracted trending keywords
│   └── user_history.pkl               ← Personalized search memory

📄 app.py                              ← Streamlit app code
📄 EDA.ipynb                           ← Data exploration + embeddings
📄 eda-ecomerce-chatbot-modeling.ipynb ← Notebook with modeling logic (MAIN WORKING CODE)
📄 requirements.txt                    ← All dependencies
📄 README.md                           ← You’re reading it!

---

💡 How It Works (End-to-End)

1. User uploads an outfit image

2. CLIP computes the embedding

3. FAISS finds visually closest products

4. LLM (Gemma-3B) takes:

   • The top match
   • Current trend strings
   • Category/style
   • And generates an outfit suggestion

5. Based on username:

   • Similar clusters are searched
   • Trend-aware + user-preferred outfits are recommended

---

📸 Sample Use Case

User Input: Image of a light pink floral blouse System Output:

✅ Top 5 visually similar items 💬 Outfit suggestion:

“Style this floral blouse with high-rise flared jeans, a straw handbag, and peach-toned wedges for a brunch-ready vibe.” 🧠 Personalized picks: Other boho-style pieces, previously liked by the user

---

💻 Installation

1. Clone this repository

git clone https://github.com/MohitGupta0123/Ecomerce_Customer_Support_Chatbot.git
cd fashion-visual-search

2. Install dependencies

pip install -r requirements.txt

Make sure you have a GPU or sufficient CPU RAM for running the LLM (Gemma-3B) and FAISS.

3. Launch the app

streamlit run app.py

---

🧰 Tech Stack

Component	Tools & Frameworks
Embeddings	OpenAI CLIP, SentenceTransformers
Vector Search	FAISS (Facebook AI Similarity Search)
LLM	Google’s Gemma-3B (quantized for 6GB GPU)
UI	Streamlit
Storage	Pickle, NumPy, Pandas
Deployment	Local (tested on 6GB RTX GPU), Hugging Face (planned)

---

🧭 Future Roadmap

• ✅ Login authentication + persistent DB for user history (SQLite)
• 🧵 Enable natural language outfit queries (NLP + filtering)
• 🎯 Filter by price, category, size, brand, or color
• 🌐 Deploy live app on Hugging Face Spaces
• 📊 Analytics dashboard (popular products, trending clusters)

---

🤝 Acknowledgements

• CLIP by OpenAI
• FAISS by Meta AI
• Gemma by Google DeepMind
• Streamlit for UI
• SentenceTransformers
• Fashion trend inspiration from Pinterest & Myntra blogs

---

📫 Contact

👨‍💻 Author: Mohit Gupta

📧 Email: [email protected]

🔗 LinkedIn

🐍 Python | 🧠 Deep Learning | 🎨 Fashion AI