Turning clinical conversations into instant summaries

A Healthcare & Life Sciences business working with Cloud Combinator on AWS. The client is anonymised at their request.

The challenge had three focus areas that needed to work together in one automated flow.

The client opens a WebSocket connection through Amazon API Gateway, which triggers a Lambda that records the connection and its status in Amazon DynamoDB. Audio is then streamed in chunks to a send-audio route, where a handler Lambda drives an AWS Step Functions state machine that runs each chunk through Amazon Transcribe, streams the transcription back to the client in real time, and stores it in Amazon S3.

Cloud Combinator delivered the pipeline against the agreed success criteria: accurate real-time transcription with Amazon Transcribe, automatic processing of appointment audio, transcript summarisation with Amazon Bedrock managed within usage limits, and robust per-appointment logging in Amazon DynamoDB with a redundancy and backup strategy across S3 and S3 Glacier, all deployed as repeatable CloudFormation stacks.

• End-to-end automation — Audio streamed from the client is transcribed, stored, aggregated and summarised without manual steps, with results returned to the clinician.
• Orchestrated and resilient — AWS Step Functions coordinate transcription and summarisation as discrete stages, making the flow easier to monitor, retry and extend.
• Traceable by design — Amazon DynamoDB tracks each connection through its full lifecycle, and both transcripts and summaries are stored durably in S3 with lifecycle rules to S3 Glacier and cross-region replication for backup.
• Deployed as code — The environment was stood up through AWS CloudFormation, so it can be rebuilt and evolved consistently as the client moves towards production.
• Handover — Cloud Combinator provided demonstration calls, Loom walkthroughs and documentation so the client's team could operate and build on the service.

This engagement proved the transcription and summarisation approach on AWS and gave the client a well-understood foundation. The next steps identified with the client, including

Additional prompt engineering, hallucination safeguards and error handling, form a clear path to harden the service to a production standard as the client scales.