Smarter Product Discovery with AI: Image Labeling, Translation, and Cross-Selling

In today’s fast-paced e-commerce landscape, customers expect smarter, more personalized shopping experiences. That’s exactly what our new project delivers—a seamless blend of AI image recognition, language translation, and cross-selling intelligence designed to help online stores better connect with shoppers.

---

🔍 Image Labeling with Google Cloud Vision

The system starts with Google Cloud Vision, which can analyze any product image and instantly detect labels. For example, uploading a kitchenware photo could return keywords like “cutting board, knife, kitchen tool.”

• Endpoint: GET /api/analyze
• Web UI: Upload an image and preview results instantly.

This not only makes search smarter but also helps stores tag and organize products automatically.

---

🌏 Instant Thai & Multilingual Translation

Since e-commerce is global, the project integrates Google Cloud Translate to break language barriers. Detected labels and queries are automatically translated into Thai (and other languages).

• Results include a description_th field.
• Customers can search in their own language.

This ensures that international shoppers feel right at home.

---

🤝 Cross-Sell Ideas with LLMs

Shopping is about discovery, not just search. Our LLM-powered engine (running locally on Ollama) generates complementary item suggestions.

• Endpoint: GET /api/related_words
• Language-aware with deterministic fallbacks
• Example: A “coffee maker” might trigger related ideas like “coffee beans, mugs, milk frother.”

This gives stores a powerful way to recommend bundles and increase cart value.

---

💬 AI-Generated Sales Pitch

Beyond lists, we also generate short, friendly sales pitches for cross-selling.

• Endpoint: GET /api/crosssell_pitch
• Optionally translated while preserving formatting
• Example: “Complete your coffee experience with freshly roasted beans and a stylish mug—perfect for your morning ritual!”

This gives marketing teams ready-to-use text snippets that feel natural and engaging.

---

🛒 Product Search & Scraping

Customers can search products using Thai queries, and the system searches through Kacee products, scraping key product details:

• Title
• Image
• Price
• Link

This allows direct integration into e-commerce catalogs.

---

🖥️ Integrated UI for Exploration

All these features come together in a clean, integrated interface (templates/index.html):

• Image preview
• Top labels (with Thai translation)
• Related products
• Raw JSON view
• LLM playground for ideas & pitches

It’s not just functional—it’s fun for both developers and end-users.

---

⚡ Graceful Degradation & Configurable Runtime

We know real-world systems must handle failures gracefully. That’s why:

• If Google/Ollama/network are down, the app soft-fails with sensible defaults.
• Config is flexible: GCP project credentials are handled via environment variables, and local Ollama models (e.g., gemma:2b) can be swapped in relatew.py.

This makes the system reliable and adaptable across different deployments.

---

🚀 Why This Matters

With this project, businesses can:

• Reduce manual tagging work
• Engage multilingual customers
• Boost sales through intelligent cross-selling
• Deliver a more delightful shopping experience

It’s a next-generation AI toolkit for e-commerce, blending cloud APIs with local LLMs for performance, reliability, and scalability.

---

✅ Next Step: We’re excited to explore how this can be rolled out to real online stores. Imagine uploading a product photo, getting instant AI insights, and boosting sales—all in one seamless flow.