The client
Arham Technosoft runs a corporate gifting business with large, spreadsheet-driven catalogues. EasyGift is the platform we built for it.
- 854products embedded
- 12vendors
- US$90–130a month to run
The challenge
Large spreadsheet catalogues: hundreds of products across dozens of categories.
Clients asked in plain language, such as “premium gifts under ₹750 for Doctors’ Day”; keyword search missed good options.
Budgets had to be honoured exactly across every suggestion and multi-item hamper.
Every enquiry needed a consistent, client-ready quote, without running any AI infrastructure in-house.
How it works
Typed in plain language, or a photo
Budget is a hard filter
One product per category, no duplicates, ranked to use the budget well
Ready to send to the client
What we did
- 01Discovery
assessed 854 products across 12 vendors, needing both text and image search; chose Amazon Bedrock over hosting models.
- 02Design
a fully serverless stack (API Gateway, Lambda, Aurora Serverless v2 with pgvector) keeping relational and vector data in one database.
- 03AI
Titan Text Embeddings V2, Titan Multimodal Embeddings G1 and Amazon Nova Lite, all through Amazon Bedrock.
- 04Build
natural-language and photo search with the budget as a hard filter; a hamper builder (one product per category, no duplicates, ranked to use the budget well); branded PDF quote export.
- 05Security
secrets from AWS Secrets Manager at runtime, a private database reachable only through RDS Proxy and an SSM bastion, protected S3 buckets, VPC endpoints instead of a NAT gateway; deployed with CloudFormation in the client’s own account in Mumbai (ap-south-1).
- 06Cost
Aurora Serverless v2 scales to its floor when idle; the NAT gateway (about US$35 a month) was replaced with VPC endpoints.
The architecture
- Amazon Bedrock
- Titan Embeddings
- Nova Lite
- Lambda
- Aurora Serverless v2
- pgvector
The result
A request typed in plain language or shown as a photo becomes ranked products, a budget-compliant hamper and a branded PDF quote in minutes.
The full environment runs at roughly US$90–130 a month.
No AI infrastructure to manage, no AI keys in the code, and no database exposed to the internet.
Want something like this for your team?
Tell us the questions your team asks or the work that slows you down. An engineer will suggest a small first build on Amazon Bedrock.
