Urdu patient intake: voice and structured history at OPD
OPD پر اردو میں مریض کی تاریخ: آواز اور منظم اندراج
20 min read
Three ways to see it
Components of the intake. A voice recorder running offline on a tablet captures the patient's complaint in Urdu or Punjabi. A speech-to-text model transcribes. An LLM rewrites the transcript into structured fields: chief complaint, duration, associated symptoms, medications, allergies, and a risk flag. The doctor sees both the original transcript and the structured note. The doctor is the source of truth, not the model.
Privacy by design. Patient voice is sensitive data under PDPA 2023. Process on device where possible, encrypt at rest, and only sync structured fields to the hospital information system if the patient consents. Show a one-screen consent in Urdu before the recording starts, with a clear opt-out that does not penalize the patient or delay care. Retain raw recordings only for as long as clinical practice requires, then delete them.
Workflow fit. The intake must not add doctor time on average; if it does, it will be abandoned within a week. Aim for a thirty-second read on the structured summary that replaces a two-minute conversation. Place the device with the OPD facilitator who already greets patients. Train two people per OPD, not one, so the workflow survives leave. Audit how often the structured summary actually shaped the consultation.
Quick check
Quick check: what makes modern AI different from a rule-based program?
The why-tree
Common pitfalls. The model paraphrases too aggressively and changes meaning. Always show the original transcript next to the structured note. The model invents symptoms not in the recording. Add a verification step where the doctor can flag hallucinations and the case is logged for retraining. The model misses local dialect terms; collect a list of fifty common Punjabi and Pashto complaint phrases from your own OPD and tune the prompt or model with them.
Try this with Claude
Mini pilot plan. Week one: set up consent flow and basic transcription on one OPD desk. Week two: integrate the structured note format with the existing chart. Week three: shadow mode where the AI runs but the doctor does not see the output, while a research assistant reviews accuracy. Week four: enable visibility to doctors. Week six: review with the clinical lead and decide whether to scale or rework.