Asking the database questions in plain English

A SaaS B2B business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas, all centred on making data insight safe, accurate and self-service.

An employee asks a question in plain English through an API, served by an AWS Lambda Function URL. AWS Lambda orchestrates the request and calls Amazon Bedrock, which uses the database schema and a knowledge base to convert the question into SQL, applying prompt engineering and guardrails to keep the response safe and on-topic.

The proof of concept demonstrated a working natural-language-to-SQL workflow in the client's AWS environment and met its success criteria. Questions asked in plain English are converted to SQL, executed against the database, and returned through an API, with a self-correcting loop, guardrails and full conversation logging in place.

• Plain-English querying — Employees ask questions in natural language and get answers from the SQL database, with no SQL knowledge needed.
• Safe, guided responses — Amazon Bedrock guardrails and prompt engineering keep answers controlled, consistent and on-topic.
• Resilient by design — A self-correcting loop fixes minor SQL errors automatically and retries within safe limits before reporting back.
• Auditable and current — Every question and response is logged in DynamoDB, and the knowledge base stays in sync with S3 automatically.
• Handover — Delivered as a CloudFormation template with documentation and a knowledge-transfer session, so the client can run and extend it.

With the pattern proven, the client can take it from proof of concept toward production, adding user authentication and wider data sources to put natural-language insight in the hands of more of the business.