Top Doctors' AI medical scribe for clinical reports - Voice AI Space Barcelona
Julien Delaunay, Researcher Innovation / AI Projects at Top Doctors España
Julian, an AI researcher specializing in NLP and explainability at Top Doctors, presents TD Listener, a medical AI scribe available to the company's roughly 90,000 doctors worldwide. He frames the problem as clinical burden: doctors spend consultations and after-hours time typing patient reports instead of focusing on the patient, which harms both care quality and the patient experience. TD Listener captures audio via the computer microphone, a phone (through a QR code), an uploaded file, or by joining telemedicine calls on Teams, Meet, or Zoom to grab the audio and generate a report. He demos the interface, showing customizable report templates (detailed first-visit versus shorter follow-up formats, easy to create new ones) and language output in Spanish, English, Italian, and Catalan. Using a sample 25-minute doctor-patient conversation, he shows the tool producing a structured SOAP report (subjective, objective, assessment, plan) with key items like medications and doses highlighted in brackets so the doctor reviews and corrects them before saving. He shares research the team is publishing comparing ASR models (Voxtral, GPT transcribe, Whisper, Google Cloud Speech-to-Text fine-tuned on clinical data) and LLMs (GPT, Gemini, open-source MedLLaMA), finding that general-purpose models outperformed expensive medical fine-tunes and that avoiding hallucination in clinical reports is paramount. The tool is already used by hundreds of doctors across Mexico, Italy, Colombia, the UK, and Spain, with planned features like auto-booking follow-ups via WhatsApp and pre-visit patient summaries. The Q&A covers GDPR and data protection (a no-data-use agreement with Azure, no audio or report retention), country-based pricing (€30 for 60 hours of audio in Spain), integration with clinics and existing software via Recall AI, the 15-person AI hub within a 500-person company, and how they manage hallucination risk by keeping doctors in the loop as verifiers rather than aiming for fully autonomous report
Recorded during a Voice AI Space Event. Check past and future events https://events.voiceaispace.com