CareDoc

Clinical documents arrive in wildly inconsistent formats: scanned PDFs, faxed images, handwritten notes, dictated audio, and structured EHR exports. CareDoc's existing OCR pipeline could only process typed text, leaving 35% of documents requiring manual data entry. Turnaround time for complete chart abstraction averaged 72 hours, causing coding backlogs and delayed reimbursements for hospital clients.

Mixpeek's multimodal pipeline processes every document type through modality-specific feature extractors: OCR with layout understanding for scanned documents, speech-to-text for dictations, and handwriting recognition for clinical notes. A taxonomy enrichment layer maps extracted content to ICD-10 and CPT codes. The unified output feeds directly into CareDoc's coding workflow, pre-populating fields that coders simply verify.

The integration replaced CareDoc's legacy OCR vendor with Mixpeek's batch processing pipeline. Documents are uploaded to a Mixpeek bucket via an S3-compatible API, processed through a collection with modality-specific extractors, and output as structured JSON. The initial deployment covered typed and scanned documents, with handwriting and audio support added in a follow-up phase over four weeks.