Executive summary
HK Infosoft runs corporate gifting operations with a catalogue sourced from multiple vendors. Customer requests such as “premium onboarding gifts under ₹2,500” rarely matched catalogue wording, so keyword search failed and staff reviewed hundreds of products by hand. DevOps TechLab built a serverless semantic search platform on Amazon Bedrock that understands intent and combines relevance with budget, vendor preference and availability.
- Client
- HK Infosoft · Corporate gifting
- Business problem
- Keyword search failed on natural-language requests, so staff reviewed hundreds of products by hand.
- What we delivered
- Semantic product search with budget-aware recommendations, on Amazon Bedrock.
- Engagement
- Discovery workshop, design, build and security, then ongoing optimisation by DevOps TechLab.
- Outcome
- Relevant products in seconds, with less manual effort and better search accuracy.
The business challenge
Customer requests did not match the way the catalogue was written, so keyword search failed. Staff reviewed hundreds of products across categories before recommending anything, budget checks were manual, and quality depended on individual experience.
Why it mattered
- Keyword search failed
Requests in customers’ own words did not match catalogue wording.
- Lengthy manual review
Staff went through hundreds of products before recommending anything.
- Manual budget checks
Every recommendation had to be checked against the budget by hand.
- Quality depended on the person
Results varied with individual experience of the catalogue.
What HK Infosoft set out to achieve
- Search the catalogue by intent, in natural language.
- Combine relevance with budget limits, vendor preference and availability.
- Cut the time staff spend reviewing products.
- Use managed AI rather than self-managed models.
- Run on a scalable, serverless base.
The solution: semantic product discovery on Amazon Bedrock
Employees type the request the way the customer said it. The platform finds products by meaning, then ranks them using budget limits, vendor preference and availability, and shows the results in a web application.
Key design choices
- Managed AI over self-managed
Amazon Bedrock with Titan Text Embeddings V2 and Amazon Nova Lite, chosen over running self-managed AI.
- Search by meaning
pgvector with HNSW indexing in Aurora PostgreSQL Serverless v2 finds products by intent, not keywords.
- Scales with demand
API Gateway, AWS Lambda and Aurora Serverless v2, with Amazon RDS Proxy handling concurrent connections.
- Fast web delivery
A React application served from Amazon S3 and Amazon CloudFront.
Security and governance
Catalogue and customer data stay inside a private AWS environment, defined entirely as code.
- Secrets managed
Credentials are kept in AWS Secrets Manager.
- Least-privilege access
AWS IAM roles control what each component can reach.
- Private database
Private subnets and Amazon RDS Proxy keep the database off the internet; VPC endpoints are recommended in place of a NAT gateway.
- Infrastructure as code
Everything is deployed with AWS CloudFormation.
Architecture on AWS
| AWS service | Role in the solution |
|---|---|
| Amazon Bedrock · Titan Text Embeddings V2 | Understands requests and products by meaning |
| Amazon Bedrock · Amazon Nova Lite | Generative AI model used alongside the semantic search |
| Amazon Aurora PostgreSQL Serverless v2 · pgvector (HNSW) | Catalogue data and semantic search |
| Amazon API Gateway · AWS Lambda | Search and recommendation logic |
| Amazon RDS Proxy | Handles concurrent connections from Lambda |
| Amazon S3 · Amazon CloudFront | Delivers the React web application |
| AWS Secrets Manager · AWS IAM · AWS CloudFormation | Secrets, access control and infrastructure as code |
How we delivered
- 01Discovery workshop
Analysed the catalogue and metadata, picked semantic search as the highest-value capability, and chose Amazon Bedrock over self-managed AI.
- 02Design
API Gateway, Lambda, Aurora Serverless v2 and Bedrock, with architecture, security and running costs reviewed before the build.
- 03Build & secure
Natural-language search with budget, vendor and availability logic, a React app on S3 and CloudFront, on private infrastructure.
- 04Optimise
Aurora scaling with demand, RDS Proxy for concurrent connections, and VPC endpoints recommended in place of a NAT gateway.
Business outcomes
Finding the right products for a request no longer means reviewing the whole catalogue by hand.
- Relevant products in seconds
Employees search in natural language and find relevant products in seconds.
- Budget built in
Recommendations combine relevance with budget limits, vendor preference and availability.
- Better search accuracy
Search understands intent, not just matching words.
- Less manual effort
Staff no longer review large catalogues by hand before recommending.
- A scalable base
A serverless platform that grows with the catalogue.
What’s next
The same platform can extend to new vendors and categories. DevOps TechLab continues to optimise search quality and running cost as HK Infosoft’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.
