Intelligent Shopping Experience powered by EDB Postgres AI & OpenShift AI
GenAI use case for Smart Search and Summarization
This solution provides personalized product recommendations in an e-commerce or retail setting, specifically for fashion items like clothing, shoes, and backpacks. By leveraging AI-driven semantic search capabilities through the AI Factory in EDB Postgres AI, running on the Red Hat OpenShift platform, and leveraging AI models served by OpenShift AI, the solution allows users to find relevant products based on providing free-form textual descriptions or by bringing custom images of products they like. It enhances the user shopping experience by presenting similar items and providing summarized insights from user reviews.
We address the challenge of users navigating large product catalogs to find items that match their specific style preferences or visual examples. It is a much more personalized experience than manually browsing or keyword-based searching, which may not capture the nuances of user intent effectively. It also synthesizes numerous user reviews into concise summaries, saving users time in evaluating products. The underlying OpenShift platform ensures the solution is scalable, manageable, and secure, while the EDB Postgres AI offers control and safety for the Sovereign AI data layer.
Key Benefits
- Enhanced Product Discovery: Users can find products that match their desired aesthetic through intuitive text or image searches.
- Personalized Recommendations: Semantic search delivers results based on the meaning and context of the query, leading to more relevant suggestions.
- Efficient Review Analysis: Automatically generated summaries of user reviews highlight key themes (e.g., Fit, Quality, Comfort), allowing for quicker assessment of product suitability.
- Improved User Engagement: A more intuitive and relevant shopping experience encourages users to explore more products.
- Scalable AI Infrastructure: Utilizing EDB Postgres AI allows for efficient management, storage, and indexing of product data, embeddings, and reviews. Leveraging the Red Hat OpenShift platform and OpenShift AI enables flexible deployment and scaling of the entire solution to accommodate customers’ operational and scalability demands, including GPU acceleration of AI processing.
- Robust Deployment & Management: OpenShift AI provides a scalable, secure, and manageable environment for the entire recommendation solution, including AI model serving and application deployment.
- Streamlined AI Operations: Elyra within OpenShift AI manages the deployment logic and tasks, simplifying the operational aspects of the AI solution.
AI Use Case Workflow
The following steps outline the use case, leveraging EDB AI Factory for embedding generation, semantic search, summary & label generation for overall key themes from a GenAI model and EDB Postgresql for data management:
- Step 1 - User Initiates Search
User Roles: End User / Shopper
- The user interacts with the search interface on the e-commerce platform.
- They can either type a textual description (e.g., "blue summer dress with floral print") or upload an image of a fashion item they like.
- Step 2 - Generating Embeddings with EDB AI Factory
User Roles: AI Engineer / Backend Developer
- The user's input, product details as text from a CSV file and images from NetApp Storage is sent to EDB AI Factory.
- The input is then converted into vector embedding by the EDB AI Factory which utilizes core AI models: A clip-based model for generating image embeddings, a new Grit-LM model for text embeddings and a Llama-3.1 model for generating product review summaries. These embeddings represent the semantic meaning or visual characteristics of the user's query.
- Step 3 - Semantic Search in EDB AI Factory
User Roles: AI Engineer / Database Administrator
- The generated embedding is used to query the EDB Postgresql database, which stores product information and pre-computed embeddings for all product descriptions and images.
- EDB AI Factory performs a semantic similarity search to find product embeddings that are closest to the user's query embedding.
- The system retrieves a list of products, ranked from most similar to least similar based on the embedding distance.
- Step 4 - Displaying Recommendations
User Roles: Frontend Developer / End User / Shopper
- The search results (ranked list of similar products) are presented to the user on the e-commerce interface.
- Each product listing includes an image, name, price, and a "Reviews" button.
- Step 5 - Accessing Product Reviews and AI Summary
User Roles: End User / Shopper / AI Engineer
- The user clicks the "Reviews" button for a specific product.
- They are directed to the product's dedicated review page.
- On this page:
- The product image is displayed.
- Below the image, a user review summary generated by a GenAI model (Llama 3 8b), running via vLLM on OpenShift AI is shown. This summary highlights key themes (Fit, Quality, Comfort, etc.) as labels derived from analyzing all user reviews for that product.
- This summary is automatically polled and updated to reflect the latest user feedback.
- Below the summary, individual user reviews are listed for detailed reading.
