Grounding AI answers in enterprise data

A FinTech business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas that the proof of concept needed to address.

The client's documents are stored in an Amazon S3 knowledge base with controlled IAM access, and indexed into an Amazon OpenSearch Service vector database so that relevant content can be found by similarity search. When an user asks a question through the web application, the request is routed via Amazon API Gateway to an AWS Lambda function that orchestrates the workflow.

Cloud Combinator delivered a functional, secure LLM application built on Amazon Bedrock, demonstrating that the client can generate grounded insight from their own data within AWS's secure environment and reduce their dependency on external platforms.

• In-house AI — The client now has a LLM application running on their own AWS infrastructure, reducing dependency on third-party platforms.
• Grounded responses — Retrieval-augmented generation bases answers on the client's knowledge base, so output is contextual and relevant rather than generic.
• Secure by design — Data access is controlled through IAM roles and policies, keeping the solution aligned with enterprise-level security needs.
• Serverless and scalable — An API Gateway, Lambda and Bedrock architecture handles requests without managing servers and scales with demand.
• Handover — Cloud Combinator delivered the working application with a demonstration and knowledge transfer so the client's developers could continue building on it.

With the RAG application proven, the natural next step (projected) is to integrate it fully into the client's product front end and extend the knowledge base across more of their data, building on the secure foundation delivered in the proof of concept.