PNJ: Enhancing Jewelry Shopping with GenAI-Powered Intent & Visual Search on AWS

Industry

Retail – Jewelry

Technology

Amazon Bedrock

Amazon OpenSearch Serverless

Amazon S3

AWS Lambda

AWS Step Functions

Amazon API Gateway

Serverless Compute

GenAI Metadata Extraction

OVERVIEW

As one of Vietnam’s most iconic jewelry brands, PNJ has built a reputation for craftsmanship, sophistication, and exceptional customer experience. With the rapid shift toward digital shopping, PNJ sought to redefine how customers discover products online—moving beyond traditional text search toward a more intuitive, image-driven, AI-powered search experience.

To achieve this vision, PNJ partnered with Renova Cloud to build a GenAI-powered Intent & Visual Search on AWS (Amazon Web Services). This modern, multimodal system empowers customers to find jewelry using natural-language descriptions, uploaded photos, or a combination of both, unlocking a new level of convenience and personalization.

KEY CHALLENGES

PNJ’s large product catalog—which continues to grow rapidly—required a search experience capable of understanding complex customer intent across multiple formats. Key challenges included:

• Limited search experience: Traditional text-based search struggled to interpret nuanced user intent such as style, material, or design features.
  
• Visual search inconsistencies: Users often uploaded images with varying lighting, angle, clarity, or background, making accurate matching difficult.
  
• Scalability needs: With millions of product variations, any solution needed to support fast, accurate retrieval at high scale.
  
• Metadata limitations: Product attributes and visual details were difficult to standardize manually, impacting search performance and consistency.
  

SOLUTION

To address these challenges, PNJ and Renova Cloud designed a production-ready, multimodal GenAI search platform capable of delivering fast, intelligent, and personalized product discovery.

1. Architecture & Deployment

The solution integrates multiple AWS-managed services into a unified search pipeline:

• Amazon Bedrock: Generates embeddings for images and text, enriches product data with AI-generated visual descriptions, and powers multimodal intent matching.
• OpenSearch Serverless (vector database): Stores and indexes embeddings for large-scale similarity search across PNJ’s extensive product catalog.
• Amazon S3: Serves as the central storage for product images and uploaded customer images.
• AWS Lambda & Step Functions: Coordinate image processing, metadata extraction, vector indexing, and multimodal similarity scoring.
• Amazon API Gateway: Provides unified APIs for mobile and web digital channels.
• Admin Portal: Allows PNJ teams to view search behavior (images + intent) and monitor product demand trends.
• Amazon DynamoDB: Storing system’s metadata for each processing phase for tracking and monitoring
  

2. Search Capabilities Delivered

• Image Search: Customers upload a jewelry photo to find similar styles instantly.
• Intent Search: Natural-language queries interpret style, material, shape, occasion, or design preferences.
• Hybrid Search: Combines image + text for best-match results (e.g., “Similar ring but in rose gold”).
• Automated Metadata Enrichment: GenAI creates standardized visual descriptions that improve catalog accuracy and downstream analytics.

BENEFITS

Financial Benefits

• Improved Product Discovery: Customers find suitable products more easily, increasing the likelihood of purchase.
• Higher Customer Engagement: Reduced search friction enhances satisfaction and long-term loyalty.
• Lower Manual Effort: Automated metadata generation significantly reduces operational overhead.

Operational Benefits

• Unified Search Experience: Consistent search capabilities across e-commerce web, mobile app, and digital storefronts.
• Scalable Architecture: Capable of supporting millions of SKUs with low-latency, high-quality retrieval.
• Actionable Insights: Admin analytics help PNJ understand customer intent and demand patterns in real time.
• Standardized Workflow: Fully automated pipeline simplifies product updates and catalog enrichment.

Performance Improvements

• Fast Response Times: Multimodal search returns results within seconds across large catalogs.
• High Retrieval Quality: AI-powered similarity search significantly improves match relevance for both images and text.
• Robust Multimodal Matching: Handles variations in lighting, angles, and camera quality while maintaining consistent search accuracy.

CONCLUSION

By partnering with Renova Cloud, PNJ has taken a step toward redefining digital jewelry shopping in Vietnam. Powered by Generative AI on AWS, the Intent Search & Visual Search Platform delivers a modern, intuitive, and highly personalized discovery experience that meets the expectations of today’s customers.

This solution not only enhances product search performance across online channels, but also lays a strong foundation for future innovations—including personalized recommendations, advanced style analytics, and fully AI-driven virtual shopping experiences. PNJ is now positioned at the forefront of retail modernization, setting a new standard for the jewelry industry in Southeast Asia.

Discover how Renova Cloud empowers organizations to build resilience, scale smarter, and run effectively with cutting-edge cloud & GenAI solutions. Explore more here