Launched this week
TeleHealth Platform
AI-assisted teleconsultation from booking to e-prescription
6 followers
AI-assisted teleconsultation from booking to e-prescription
6 followers
AI-Powered Teleconsultation Platform Patients face fragmented booking and follow-up across channels, with weak continuity of care records. Clinicians spend significant time on documentation during and after consults, reducing throughput and quality of attention. India-specific requirements—DPDP consent, Telemedicine Practice Guidelines, PHI handling—must be designed in, not bolted on. AI in healthcare must augment clinicians without bypassing accountability for diagnosis or prescriptions.
How does the platform handle the AI scribing during consults when the patient speaks in mixed Hindi-English or regional languages common in India?
@bulem234983 Right now, scribing runs on Whisper (transcription) → GPT-4o (SOAP note drafting). Whisper has some built-in multilingual support, so it partially handles Hindi and Hindi-English mixing already, but I haven't added anything specific for it yet, like language detection or normalizing mixed-script text before it hits GPT-4o. That's a real gap given how common code-switching is in Indian consults.
I am planning to fix it, and thinking to focus on three things:
Detect language per audio chunk instead of assuming English, so transcription doesn't default to poor guesses.
Clean up the transcript before it reaches GPT-4o Whisper often mixes Hindi written in Roman script with English inconsistently, so normalizing that first would help.
Prompt GPT-4o to expect code-switching tell it to treat Hindi-English mixing as normal, not something to 'fix' into pure English.
For now, the safety net is that clinicians should always review and edit the AI draft before it's finalized, so any transcription mix-ups get caught before they reach the patient record.