Overview

myDORA is a digital operating-room platform covering the full lifecycle of a surgical procedure. It manages patient records, documents and conditions; schedules procedures; records endoscopic video; and turns completed procedures into transcripts, structured summaries, clinical documentation and teaching material.

Before a procedure, a surgical team can record its goals and plan, assign attendees and complete WHO-style safety checklists. During surgery, myDORA captures HD video and supports the team without requiring keyboard interaction. Afterwards, the team can review the recording, outcomes and generated material or share a live view with remote observers.

The platform operates in European hospitals and is designed around GDPR and EHDS requirements. Access is auditable, and the system supports DICOM for hospital imaging workflows and SNOMED for structured clinical coding.

My work

I architected and led the development of myDORA for SofMedica Ventures, while building most of the system myself. Over the course of the project I hired and led several remote developers and a designer. My work spans the product, application, AI and media pipelines, clinical integrations and production infrastructure.

AI in the clinical workflow

The voice assistant uses an on-device wake word followed by medical speech recognition. It can create notes and bookmarks, open patient documents, start a procedure and analyse DICOM images. Its responses are grounded in the active patient and procedure, with semantic search over that patient's documents rather than a general-purpose chat interface.

After a procedure, the recorded audio is transcribed with speaker separation. AI then produces a structured summary and draft clinical material for review and editing. Presentation decks are generated from the procedure record, combining its diagnosis, goals, notes, steps, bookmarks and key moments. DICOM analysis can also identify and highlight relevant regions in medical images for educational use.

Recording through unreliable networks

Unreliable network infrastructure is a common reality in the southern and eastern European hospitals where myDORA runs. Connections regularly degrade, disappear during a procedure and return later. A recording cannot be recreated, so the media pipeline is built around interrupted connections rather than treating them as exceptional. Recording progress is monitored continuously; stalled sessions are restarted, and damaged playlists can be reconstructed from the video segments that reached storage.

LiveKit writes segmented recordings to S3 before downstream processing begins. Audio extraction, transcription and document generation happen as separate stages, allowing each to be retried without recapturing the procedure.

Cost at surgical recording volumes

A surgeon records about thirty hours per month on average, and a single operation can last more than ten hours. Minimising cost per surgeon was therefore a design goal rather than a later optimisation. The first version used Daily.co and Mux, but repeated video processing became too expensive at these volumes and vendor-side failures left too little control over the critical recording path.

The replacement pipeline stores segmented video as its primary format and extracts the much smaller audio stream for transcription. MP4 conversion is performed only when someone requests a download, while different AI models are used according to the cost and capability required by each task.

The move to LiveKit and AWS was completed while the platform remained in use. The frontend was later rebuilt in the same way: the second generation ran beside the first on a shared backend, allowing users to move over without a big-bang migration or service interruption.

Today

myDORA has been running continuously for about two years and is live in hospitals across Europe. It is used for patient and procedure management, surgical recording, documentation, teaching and remote observation.