Executive summary
Spotem’s corporate gifting customers often enquire with photos, screenshots or reference images rather than product names. Staff compared those images against hundreds of catalogue products by hand, and results depended on who was searching. DevOps TechLab built a serverless visual search platform on Amazon Bedrock with Amazon Nova Lite and Titan Multimodal Embeddings, so staff upload the image and receive visually similar products within seconds.
- Client
- Spotem · Corporate gifting
- Business problem
- Photo-based enquiries were matched against hundreds of products by hand.
- What we delivered
- AI visual product search on Amazon Bedrock: image analysis, multimodal embeddings and vector search.
- Engagement
- Discovery, design, build and security, then ongoing optimisation by DevOps TechLab.
- Outcome
- Visually similar products within seconds, without depending on one person’s memory of the catalogue.
The business challenge
Many of Spotem’s enquiries arrived as images rather than product names. Staff had to compare each image against hundreds of products manually, keyword search could not find visually similar items, and the quality of results depended on who was searching.
Why it mattered
- Slow responses to image enquiries
Each photo meant a manual search through hundreds of products.
- Keyword search can’t see
Text search could not find products that simply looked like the customer’s reference.
- Results depended on the person
Recommendations relied on individual memory of the catalogue.
- No appetite for computer vision infrastructure
The business wanted image AI without running its own vision models.
What Spotem set out to achieve
- Accept a customer’s photo or screenshot as the search.
- Return visually similar products from the whole catalogue in seconds.
- Make results consistent, whoever is searching.
- Use managed AI, with no computer-vision infrastructure to run.
- Keep images and data secure.
The solution: visual product search on Amazon Bedrock
Staff upload the customer’s reference image. Amazon Nova Lite analyses it, Titan Multimodal Embeddings turn it into a vector, and vector search finds the closest products in the catalogue.
- 1Upload
Staff upload the customer’s image securely
- 2Analyse
Amazon Nova Lite analyses the image
- 3Embed
Titan Multimodal Embeddings turn it into a vector
- 4Match
pgvector with HNSW finds the closest products
Key design choices
- Managed image AI
Amazon Bedrock with Amazon Nova Lite and Titan Multimodal Embeddings, so there is no computer-vision infrastructure to manage.
- Fast vector search
pgvector with HNSW indexing in Aurora PostgreSQL Serverless v2 finds the closest products quickly.
- Fully serverless
Amazon API Gateway, AWS Lambda and Amazon S3 scale with demand.
- Lower networking cost
VPC endpoints reduce networking cost and keep traffic private.
Security and governance
Customer images and catalogue data stay inside a private AWS environment, defined entirely as code.
- Secure image upload
Images are uploaded securely to Amazon S3.
- Least-privilege access
AWS IAM roles and AWS Secrets Manager control access and credentials.
- Private database
The database sits on private networking behind Amazon RDS Proxy, with VPC endpoints for AWS services.
- Infrastructure as code
Everything is provisioned with AWS CloudFormation.
Architecture on AWS
| AWS service | Role in the solution |
|---|---|
| Amazon Bedrock · Amazon Nova Lite | Analyses the customer’s image |
| Amazon Bedrock · Titan Multimodal Embeddings | Turns images and products into comparable vectors |
| Amazon Aurora PostgreSQL Serverless v2 · pgvector (HNSW) | Catalogue data and fast similarity search |
| Amazon API Gateway · AWS Lambda · Amazon S3 | Serverless application and secure image storage |
| AWS IAM · AWS Secrets Manager · Amazon RDS Proxy · VPC endpoints | Access control, secrets and private networking |
| AWS CloudFormation | Infrastructure as code |
How we delivered
- 01Discover
Studied how image-based enquiries arrive and chose Amazon Bedrock with Nova Lite and Titan Multimodal Embeddings.
- 02Design
A serverless architecture with vector search on pgvector and HNSW indexing.
- 03Build & secure
Secure image upload, AI image analysis, multimodal embeddings and a recommendation service, on private infrastructure.
- 04Optimise
Faster search with HNSW, Aurora tuned for workload-based scaling and lower networking cost.
Business outcomes
Photo enquiries that used to mean a manual catalogue search are now answered with matching products within seconds.
- Matches within seconds
Staff upload a reference image and get visually similar products within seconds.
- Consistent recommendations
Results no longer depend on one person’s memory of the catalogue.
- Faster replies to customers
Image-based enquiries are answered without a manual search.
- Nothing to host
Managed AI on Amazon Bedrock, with no vision models to run.
- Secure by design
Private networking, managed secrets and infrastructure as code.
What’s next
The same platform can extend to new catalogue categories and vendors. DevOps TechLab continues to tune search speed and cost as Spotem’s catalogue grows.
DevOps TechLab is an AWS Advanced Tier Services Partner with the AWS Generative AI Competency and AWS SMB Competency. We design, build and run cloud and generative AI solutions on AWS for growing businesses, with 265+ cloud environments under management.
Want AI like this working for your sales team?
Tell us where quotes, searches or proposals slow your team down. We will suggest a small first build on Amazon Bedrock, with the running cost estimated up front.
