4 demos, one family of models
Open models for Indian clinical records — sorting documents, reading them, structuring them, stripping the parts that identify a patient, and transcribing what was said.
Document classifier
Name the document from a single page in about 30 ms — one of 27 classes, plus whether it is medical, whether it is handwritten, and how legible the scan is.
PII redaction
Find personal information in a document image or in free text, then redact, pseudonymise or anonymise it — with the detections shown before anything is removed.
Medical records
Upload a clinical document. It is classified, parsed into structured entities, codified to SNOMED / LOINC / ICD-10-CM and assembled into a FHIR R5 bundle.
Speech to text
Record a clinical consultation or upload a clip of up to five minutes, and get the transcript back — chunked and transcribed server-side.
Documents are processed in memory and discarded shortly afterwards. Nothing is retained.
This runs on a single small GPU, for demonstration only — please go gentle on it. For production workloads, see our production APIs.