DeepBrief Sign in See DeepBrief in action

AI platform for automated
performance analysis
in clinical training

Expert-level feedback.
Every practice.
Anywhere.

DeepBrief gives every trainee expert-level feedback on every practice — and gives the institution, for the first time, a clear picture of who is ready and what to teach next.

Built by physicians. Developed with leading medical centers and academic institutions.

See DeepBrief in action
A phone on a tripod recording a neonatal resuscitation simulation

Partner institutions

The Hebrew University of Jerusalem Hadassah Medical Organization Magen David Adom — Israel's National EMS

Your standard, loaded in first

Your protocols, station scripts, rubrics and performance criteria are loaded into the system before anything is recorded. The system measures against your standard — not a generic one.

1Capture

A trainee, an instructor or a full team records the session — in a simulation centre, a skills lab, a classroom, a department, or at home. Any browser. The audio and video sources you already use.

2Analyze

Speech and action are transcribed, attributed to roles, and placed on a single verified timeline. Each criterion is scored against your rubric, with a reference to the exact moment it was met, partly met or omitted.

3Improve

The participants receive a structured performance analysis within minutes. The institution receives the same material as accumulating data — by cohort, by station, by criterion.

Measured against expert examiners.

In a prospective, blinded study

More accurate

than the instructor in the room, measured against blinded expert examiners.

Across 372 checklist items from 15 recorded simulation encounters, exact agreement with the expert reference standard was 70.2% for DeepBrief and 63.4% for the instructor scoring in real time.

Prospective, single-blinded agreement study. Hadassah Ein Kerem · Hebrew University of Jerusalem · IRB #26112025.

96.5%

DeepBrief agreement within one ordinal category

372

Item-level observations · 15 encounters

2

High-fidelity scenarios · blinded expert reference standard

Automated Versus Instructor Checklist Scoring in Emergency Medicine Simulation: A Prospective Agreement Study

Podium presentation, ACEP 2026, Chicago

Design

Prospective, single-blinded agreement study in the emergency medicine residency at Hadassah Ein Kerem Medical Center, affiliated with the Hebrew University of Jerusalem. Fifteen recorded encounters across two 20-minute high-fidelity scenarios — trauma with methanol poisoning, and an unconscious patient with carbon monoxide toxicity. Checklists were written in the national board examination format and revised by two emergency medicine specialists who serve as examiners for the Israeli Association of Emergency Medicine. Two blinded, board-certified emergency physicians reviewed every recording to set the reference standard. Approved by the Helsinki Committee, Faculty of Medicine, Hebrew University of Jerusalem (IRB #26112025).

Result

70.2%

DeepBrief

63.4%

Instructor, in real time

Exact ordinal agreement with the expert reference standard, across 372 scored items. Agreement within one ordinal category was 96.5% and 94.4%.

Weighted kappa

Favoured automated scoring in both scenarios: 0.38 (95% CI 0.22–0.50) vs 0.28 (0.14–0.41) in Case 1, and 0.79 (0.68–0.87) vs 0.65 (0.50–0.78) in Case 2. All p < 0.001.

Forthcoming

Automated Detection of Missed Checklist Actions in Emergency Medicine Simulation

IMSH 2027, New Orleans.

Built for institutions that teach, certify and train.

Medical and nursing faculties

A faculty is the only place with a long relationship with the learner — years, not a course — and the only place where progress is worth tracking as a curve rather than a result. Cohorts grow while instructor numbers stay flat, which is why practice gets rationed first.

Every student gets a structured analysis of every practice, including practice done outside the faculty. Every head of course gets the cohort picture: which criteria are being missed, by whom, and what to teach next. Candidates arrive at the exam prepared, and examiners spend their time on what only they can do.

For faculties →

EMS and paramedic academies

A body that certifies people is a school, and it trains at a scale most faculties do not reach. Sessions run in vehicles, in the field and in classrooms rather than in a fixed room, which makes anything that depends on installed infrastructure unusable.

Trainees record their own sessions on a device they already carry and receive an analysis against the national protocol. The academy gets course-wide and cohort-wide reporting from material it never had before.

For EMS academies →

Simulation centres

The scarce resource in a simulation centre is not rooms — it is the instructor who has to run the scenario, tag events while it happens, then lead a good analysis afterwards under time pressure. Quality falls across a long day, and no two instructors deliver the same session.

DeepBrief produces the analysis itself, so the instructor arrives at the conversation with the timeline already built and their attention free. It requires no change to the AV you already run and no integration project.

For simulation centres →

Hospital departments

Practice in a department competes directly with clinical work, so it happens in short cases at the bedside rather than in scheduled sessions, and it is almost never reviewed afterwards. Expertise is available for minutes at a time, and residents rotate out before anyone has formed a view of them.

Short cases run where the work happens and are analyzed against the department's own protocols. Performance is tracked across a rotation instead of recalled at the end of it.

For departments →

Build better clinicians.

The limit on clinical training has never been the willingness to practice. It is that every practice needs someone qualified watching it, remembering it, and saying something useful afterwards. That expert is finite — so practice gets rationed, and the feedback that does happen depends on who was in the room and how late in the day it was.

DeepBrief removes the rationing. A student, a resident, a paramedic or a full team can practice again and receive a structured analysis every time — which is what deliberate practice has always required and clinical training has never been able to supply.

More practice, without more expert hours.

Cohorts and caseloads grow; senior clinicians and instructors do not. They are not replaced — their time is spent where only they add value: the nuance, the judgment, the conversation.

A picture across sessions, for the first time.

Not an impression of how a cohort, a course or a department is doing, but which criteria are being missed, by whom, and over what period.

One consistent standard.

The same criteria, applied the same way, to every trainee, by every instructor, at every hour of a long training day.

If it has a standard, it can be measured.

DeepBrief does not assume one kind of session. It reads whatever format your standard was written for.

High-fidelity team simulation

Resuscitation, trauma, neonatal, obstetric and paediatric emergencies. Speech is attributed to roles — team leader, team members, patient — and commands, confirmations and clinical actions are timestamped. Each item of the protocol, whether ACLS, NRP or your own version of it, is scored as complete, partial or omitted with a reference to the moment it happened. The result is one agreed account of the scenario, so the analysis starts from what happened rather than from what anyone remembers.

OSCE and structured stations

The same rubric applied identically at every station, in every room, on every day of the cycle. Each encounter is scored against the checklist the department wrote, item by item. Station-level reporting also shows whether the station itself ran as defined — whether the required information was delivered, whether the scenario was diverted. Used to prepare candidates, and alongside examiners as a second opinion; never as the sole examiner.

Communication skills

History taking, patient education, breaking bad news, informed consent, shift handover, case presentation. These sessions need a microphone and nothing else, which means they can run in a classroom, in a corridor, or between two trainees with no room booked at all — the lowest-friction way to put deliberate practice in front of an entire cohort.

Procedural and physical examination

This is the hardest part of the problem, and it is where DeepBrief is strongest. The video engine reads physical action directly from the recording — exposure, glove and gown, the palpation and auscultation sequence, pulse-check timing, compression pauses, sharps handling — and places every action on the same verified timeline as the speech, so a rubric item is scored on what was done, not only on what was said.

See it run on your own standard.

A short session on your own material — your rubric, one of your stations, your protocol — analyzed in front of you.

A faculty, an academy, a simulation centre, a department. If you are the person who would bring this in, this is the right form.

Sending…

We opened your email client instead — send the message and we will have your details.

Thank you — we will be in touch.

A member of the founding team will reply personally.