Administrator sign-in
Clinical accuracy
F1 by entity type
How it is measured
Each consultation is synthesised with Indian neural voices plus ambient room noise, run through the full pipeline, and the generated note is compared with a clinician-written reference: diagnoses (ICD-10 category), medicines (name, strength, frequency, duration), investigations, vitals and allergies. Unseen-case rounds are the first run on consultations never used while tuning — the honest estimate of accuracy on new patients (EoI target: entity F1 ≥ 92 %). After each round those cases join the tuning set and new unseen cases are written. WER/CER compare the transcript with the spoken script.
Language distribution sessions by detected language
Models & node
Audit chain
Recomputes every hash from the first event; any altered or removed row breaks the chain.