Gallery
Description
A brief explanation of what the project does and what makes it unique
Emergency physicians lose a large part of the visit to documentation, and much of what they retype is already on the episode: triage priority, reason for admission, allergies, orders, and results. TrueFlow now closes that gap from the tracking board. The physician opens documentation on the current visit, scans a secure QR code on a phone, sees the episode context, dictates the assessment in European Portuguese, and receives a structured SOAP draft. They review and edit it, then send the signed note back to the same episode.
What makes it unique is that the note is tied to the live TrueFlow visit. The phone never receives a chart inside the QR code. The code carries only a short-lived token, and the clinical context is loaded on the server for that episode alone. Speech is captured in Portuguese as spoken in clinical practice. The draft draws on the dictation and on this visit, and it returns to the board only after the physician has reviewed, corrected, and accepted it.
The role of AI in the solution
AI works in two steps, and it stays in a drafting role.
First, the phone turns the physician's spoken assessment into text with speech recognition set to European Portuguese. The physician can correct the transcript before anything is generated, and can type the assessment if the microphone is unavailable.
Second, a language model reads that transcript together with the current episode: identity useful to the note, Manchester priority, presenting complaint, allergies, background for this visit, diagnoses, and orders with any results already back. It writes a SOAP proposal in Portuguese: what the patient reports in Subjective, factual findings in Objective, a cautious working impression in Assessment, and concrete next steps in Plan. Pending tests stay pending. Vitals, doses, and diagnoses that were never dictated and are not on the episode are left out. If the dictation and the record disagree, the dictation leads. The model returns an editable draft. The physician confirms what is stored.
Summary of the reached solution
The two-day prototype runs end to end on TrueFlow.
From the emergency board, a physician starts documentation for an open visit and gets a QR code. The code expires, works for one phone at a time, and contains no clinical data. Scanning it opens a mobile page with the context of that episode only. Earlier visits are left out on purpose, so the note documents this attendance.
On the phone, the physician dictates the assessment in European Portuguese, or types it, then asks for a SOAP note. The draft comes back in four editable sections, written as an emergency note: a short opening from the episode, symptoms from the dictation, objective facts without invented results, a hedged working impression, and an actionable plan. The physician edits any section and sends the reviewed note back.
On the board, the episode shows that a reviewed SOAP note is available, including when it was returned. It can also be removed from the visit. The loop that the hackathon asked for is in place: select the episode, scan, dictate with the chart already in view, review the AI draft, and return it to the clinical workflow with the physician still responsible for the record.