Client: Spotem

The Challenge

Spotem’s corporate gifting business frequently received customer enquiries accompanied by photographs, screenshots, or reference images rather than detailed product descriptions. Employees had to manually compare uploaded images against hundreds of catalogue products, making product discovery slow, inconsistent, and highly dependent on individual experience.

Traditional keyword searches were unable to interpret visual product characteristics, resulting in longer response times and missed opportunities to recommend suitable alternatives.

– Customers often submitted images instead of product names or detailed descriptions.
– Manual comparison of reference images against catalogue products consumed valuable operational time.
– Keyword-based search could not identify visually similar products.
– Product recommendations depended heavily on employee familiarity with the catalogue.
– Spotem wanted enterprise AI capabilities without managing computer vision infrastructure or machine learning models.

The primary objective was to build an AI-powered visual product search platform capable of understanding customer images, identifying similar products, and delivering intelligent recommendations through a secure AWS-native architecture.

Discovery

DevOps TechLab worked closely with Spotem’s business and technical stakeholders to evaluate how artificial intelligence could improve visual product discovery.

– Analysed existing customer enquiry workflows involving reference photographs and product images.
– Evaluated image recognition approaches and selected Amazon Bedrock to eliminate AI infrastructure management.
– Identified Amazon Nova Lite and Titan Multimodal Embeddings as the preferred services for multimodal image understanding.
– Defined requirements for image upload, visual similarity search, intelligent recommendations, and secure image storage.
– Established AWS security requirements including private networking, IAM-only access, encrypted secrets management, and Infrastructure as Code.

The discovery phase established a scalable AWS architecture capable of delivering intelligent visual search with minimal operational overhead.

Onboarding

Based on discovery findings, DevOps TechLab designed a fully managed AWS architecture focused on image understanding and visual product recommendations.

– Designed a serverless architecture using Amazon API Gateway, AWS Lambda, Amazon S3, Amazon Aurora PostgreSQL Serverless v2, and Amazon Bedrock.
– Selected Amazon Nova Lite for image understanding and Titan Multimodal Embeddings for visual similarity search.
– Designed a vector search architecture using pgvector and HNSW indexing.
– Validated architecture, security controls, and deployment strategy with Spotem before implementation.
Planned phased implementation to ensure smooth rollout.

Operations & Support

– Built secure image upload workflows using Amazon S3.
– Implemented AI-powered image analysis using Amazon Nova Lite.
– Generated multimodal embeddings for visual similarity search.
– Indexed product embeddings using pgvector within Amazon Aurora PostgreSQL.
– Built intelligent recommendation services using AWS Lambda.
– Secured the environment using IAM, AWS Secrets Manager, Amazon RDS Proxy, private networking, and VPC endpoints.
– Provisioned the complete environment using AWS CloudFormation.

Each deployment stage was validated with Spotem before production rollout.

Optimisation & advisory

– Optimised multimodal search performance using HNSW indexing.
– Tuned Aurora Serverless v2 for workload-based scaling.
– Reduced networking costs using VPC endpoints.
– Recommended Amazon Cognito for future user management.
– Defined a roadmap for expanding AI-assisted recommendation capabilities.

Summary

By combining Amazon Bedrock, Amazon Nova Lite, Titan Multimodal Embeddings, AWS Lambda, and Aurora PostgreSQL Serverless v2, DevOps TechLab helped Spotem transform image-based customer enquiries into intelligent product recommendations. Employees can now upload customer reference images and receive visually similar product suggestions within seconds through a secure, scalable, serverless AWS architecture.

About DevOps TechLab

DevOps TechLab helps organisations design, build, and operate cloud-native and AI-powered solutions on AWS. Our expertise spans serverless architectures, Amazon Bedrock, generative AI, data engineering, DevSecOps, and AWS Well-Architected best practices, enabling customers to modernise applications, accelerate innovation, and deliver secure, scalable cloud solutions with confidence.