Rhythm Healthcare, Jul 2023 to now
EMR ingestion and AI triage on AWS Bedrock
An ingestion endpoint for EMR messages, built end to end, with every incoming report triaged automatically by a model on AWS Bedrock.
10,000+
HL7/XML transmissions triaged a day
- Client EMR systemsmessages through Redox
- Ingestion endpointevery message variant
- Amazon SQS10,000+ a day
- Triage on Bedrockevery report
- Clinician flagcritical issues surfaced
Audit logevery decision recorded
The problem
Rhythm’s platform connects to client EMR and EHR systems. Messages arrive through Redox in many real-world variations, and the reports that point to a critical patient issue need a clinician’s attention fast.
Stack
AWS Bedrock, Amazon SQS, Redox, HL7/XML, NestJS, React
What I built
- 1Built the EMR ingestion endpoint end to end for Redox messages, covering the message and use-case variations that show up in production.
- 2Implemented the AI triage flow, so every incoming report is triaged automatically by a Bedrock-based model.
- 3Runs 10,000+ HL7/XML transmissions a day through SQS, with full audit logging.
- 4Shipped the clinician features on top, frontend to API: the transmission report mini-menu and sidebar, a rebuilt Sign-all for signing every report at once, and multiple diagnosis codes per patient.
- 5Cleaned up a client’s EMR data that wasn’t in a reliable state, validating it so the workflows behaved consistently.
Why I built it this way
Every report, not a sample
Triage runs on every incoming report, so nothing critical waits for someone to open it first.
Handle each variation explicitly
Real EMR messages branch in many directions. Covering each path on its own kept new variants from breaking the ones already in production.
Record every decision
In a HIPAA setting, every triage decision lands in an audit log, so any flag can be traced back to the message behind it.
“One of his most impactful contributions was building the backend infrastructure for our clinical support AI product. This system leveraged AWS Bedrock and Anthropic’s Claude models to flag critical patient health issues requiring attention… I would welcome the opportunity to work with Ashfaq again and recommend him without reservation.”