Turning member handbooks into instant answers

A SaaS B2B 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.

An administrator uploads a member handbook to an Amazon S3 bucket. That upload triggers an AWS Lambda function that ingests the document into an Amazon Bedrock Knowledge Base, where it is converted to embeddings and stored in an Amazon OpenSearch vector store. Keeping ingestion event-driven means the assistant always reflects the latest version of each handbook with no manual re-indexing.

The proof of concept met its acceptance criteria, demonstrating that Amazon Bedrock Knowledge Bases are a suitable foundation for a member-facing handbook assistant and giving the client a working reference for a future production build.

• Grounded answers — Members can query handbook content in natural language and receive responses drawn directly from the source documents.
• Automatic freshness — A new or updated handbook dropped into Amazon S3 re-syncs the Knowledge Base automatically, so answers stay current with no manual re-indexing.
• Clean escalation — The assistant recognises when a member needs a person and can route the request to the campus manager, with booking and lead capture designed to flow into the client's HubSpot CRM.
• Safe by design — A guardrail check on incoming prompts, plus Amazon Cognito authentication on the front end, keeps the assistant scoped to legitimate member use.
• Handover — Cloud Combinator delivered the working solution with a demonstration, supporting materials and documentation so the client's team could evaluate a production rollout.

With the approach validated, the natural next step (projected) is a fully branded production deployment in the client's own AWS account, extending the assistant across every campus and completing the live email and SMS escalation and HubSpot integration proven in the proof of concept.