# Trainer - Capturing Lost Data in Medical Simulations.mp4
So every single day in medical simulation labs, lives are, you know, theoretically saved,
right? Theoretically.
And every single day, the exact data of how those lives were saved or um how they were lost is just instantly thrown in the trash.
It's completely gone.
Exactly. As a training program manager or an instructor running these highstakes programs, you know that your most valuable asset is just evaporating into thin air. The second the scenario ends.
It is a I mean it's a profoundly frustrating reality for educators. You know, the scenario powers down, the adrenaline fades, and all that rich subtle context just vanishes before the debrief even begins.
Yeah. The the exact second a paramedic student realized the airway was compromised or
right the specific tone of voice they use when they finally asked for help or that chaotic overlapping communication. It all disappears
and you were just left relying entirely on your own. memory or maybe like some quickly scribbled notes to reconstruct an incredibly complex multi-person human interaction
which is almost impossible to do accurately.
So we are taking a deep dive today into a stack of internal white papers and brand guidelines from a company called Vecommed.
Yes, very interesting stuff.
Our mission today is to look at how they are trying to fundamentally change that dynamic for you. And um what stands out immediately is what they call their V0.5 semantic res. it.
Yeah, that semantic reset is essentially a complete overhaul of the vocabulary they use to describe their own technology and their mission really.
Okay, let's unpack this because they are explicitly stripping away the typical, you know, Silicon Valley tech bro jargon
completely. You won't find them talking about autonomous AI doctors or replacing human judgment with algorithms.
Thank goodness,
right? Instead, they are pivoting to define themselves strictly as a training first learning company,
which which is frankly a massive departure from the standard pitch in this industry. I mean, usually when a company develops medical AI, their ultimate goal is to put it in the back of an ambulance.
They want it live.
Exactly. They want it in the field saving the day. But Veco looked at the live prehosp environment and essentially decided to just walk away from it, at least for now.
Yeah, they did.
So, why abandon the actual highstakes world of live emergency medicine to focus exclusively on your simulation lab?
Well, it comes down to a very So, understanding of how emergency medicine actually functions in reality. They detail this in white paper 02.
Okay, what do they say?
They point out that emergency medicine is performed in motion. It's loud. The environment is totally uncontrolled and you know the stakes are literal life and death.
No, it's chaos.
Total chaos. So, if you are trying to train an intelligent software system, relying on live ambulance runs is incredibly slow and highly risky. The encounters just aren't repeatable.
You certainly cannot safely test the boundaries of a medical protocol in a real patient just to see what happens.
Exactly. But a training program is the complete opposite of that. Your lab is the ultimate proving ground
because the encounters happen every single day.
Right. They are highly frequent and most importantly they are completely repeatable. You can take the exact same pediatric asthma protocol and run it 30 times.
And you can deliberately manipulate the variables, right? Like crank up the background noise, swap out a senior student for a junior one,
introduce a catastrophic equipment failure right in the middle of a procedure. All of that. And crucially, error can be studied safely.
Safely. Yes.
That is the engine of their entire model. Veimemed's technical lineage actually stems from an earlier prototype called Voxil. Vox XEL.
Oh, right. The edge processing system.
Exactly. That system was built for edge processing, meaning all the computing happened locally on the device itself rather than um sending data to a distant cloud,
which you need in a chaotic environment.
Right. It was designed to capture speech and structure context in highly chaotic environments. Now they are taking that robust edge processing architecture and applying it specifically to capture the structured chaos of your training scenarios.
You know the analogy that comes to mind is an elite athletes film room but specifically the difference between game tape and practice tape.
Oh that's a great way to look at it.
Like if I want to understand how a quarterback handles a highly complex defensive blitz. Game tape is okay but it's messy and full of variables I can't control. Right.
I want to study the practice tape where the defense ran that exact same blitz 20 times in a row, deliberately stress testing the quarterback's reactions under controlled conditions.
That isolates the variables perfectly. But to extend your analogy, game tape is purely visual. What Voxil and this new iteration of the technology do is capture the auditory and state-based reality of the room.
Meaning what exactly?
It's analyzing the overlapping voices, the timing of the monitor alarms, the uh hesitation before a command is given.
Okay, but let me challenge the core premise here though. If I am building a sophisticated machine learning model, I wanted to learn from the absolute best.
Sure, that's a standard thinking,
right? But learners are by definition noviceses. They are making a massive mess of things. Why would Veco build an entire architectural framework based on the data of people who don't actually know what they are doing yet?
It's a fair question.
When the log path be to feed the system pristine expert data to teach it what perfect performance looks like?
Well, Veco's core philosophy directly addresses this and it's actually a brilliant reframing of how competence is built. Their documents argue that while experts show what excellent performance looks like, experts mostly just show you the finish line.
Learners, on the other hand, reveal the transitions.
Tell me more about what they mean by transitions. What transitions?
So, a database of flawless expert executions doesn't really help an instructor teach a struggling student how to get to that level.
Right. It just shows the end result.
Exactly. But the learners show the messy, nonlinear path to competence. They show what a novice medically notices first when they walk into a room.
They reveal the exact measurable moment when cognitive overload sets in and communication completely breaks down. They show the specific psychological and tactical steps a team takes to recover from, say, a dropped handoff.
that path, the mistakes, the corrections, the gradual building of situational awareness, that is the exact empirical data you need to systematically improve a curriculum. That reframes the entire purpose of the sim lab. Honestly, the learners aren't just, you know, flawed versions of experts. Their specific mistakes are the actual raw material needed to map out better teaching strategies.
But practically speaking, how are we capturing those subtle mistakes? Because Nobody wants to outfit their simulation lab with a million invasive cameras and microphones, turning the whole place into a rigid surveillance state.
And Veco is highly aware of that friction, which is why white papers 03 and 07 lay out a very specific operating model called the encounter to capability loop.
The encounter to capability loop.
It's structured in four distinct stages. Capture, learn, improve, and act.
Let's break those down. Starting with capture, because from reading the source material, quote unquote, Capture is not just hitting record on a GoPro.
No, not at all. It is selective observation. The system is capturing specific data streams, audio logs, precise timestamps, and the shifting states of the scenario itself.
It is not indiscriminate recording.
Right? The goal of the capture phase is purely to gather the raw inputs necessary to assist the instructor with afteraction reviews or to reconstruct the timeline of a simulated patient handoff.
Here's where it gets really interesting to me, though. The learn and improve of parts of that loop. Specifically, how the software handles its own limitations.
Yes, the invariance,
right? They have embedded what they call safety and failure invariance. The system fundamentally assumes it will be wrong sometimes, which is honestly incredibly rare for a tech company to just admit.
It is a necessary admission when dealing with the realities of emergency medicine. One of their core invariants requires that uncertainty must be made visible.
meaning if a student bumps a microphone and the audio trans description is garbled or if a sensor on the mannequin misses a chest compression because of a weird angle.
The system isn't just going to guess.
Exactly. It is not allowed to guess or hallucinate a result. It explicitly flags the missing data for the instructor.
And it pairs that with something called a nonblocking fallback. This is huge for anyone running a lab.
If the facil's network completely crashes or the software suddenly glitches in the middle of a massive final exam scenario, the scenario doesn't stop.
Not at all.
The software doesn't freeze. is the mannequin. Your lab keeps running exactly as it did before you ever installed the technology.
The technology degrades gracefully. It is designed to be a completely passive observer until you need it.
I still have a major concern about the psychology of the student in that room though.
Oh, the performance anxiety aspect.
Yeah. If you introduce software that is logging their every word and timestamping their actions, do they start like performing for the tech? Do you get students weirdly narrating their actions? saying things like, "I am now applying the tourniquet to the left leg."
Exactly. Just to make sure the algorithm registers it instead of focusing on the actual bleeding patient in front of them.
That is the exact trap this architecture is meant to avoid. Vecommed draws a very strict boundary in their design principles. They state the product should fit training, not make training fit the product.
So the learner shouldn't even know the system is running.
Effectively, yes. The learner's interface is the mannequin, their gear, and their team.
Okay, that makes sense.
Furthermore, the documents heavily emphasize protecting your authority as the instructor. The software is simply a control surface for your review.
A control surface,
right? It organizes the timeline and structures the data so you can quickly say, "Let's look at minute 4. This is where the medication error occurred."
It's basically a teaching assistant, like a TA who takes meticulous notes, but is humble enough to highlight what they couldn't hear rather than some arrogant robot trying to grade the students.
Exactly. The software does not objectively grade the student. It does not declare a pass or a fail. You are the sole authority for evaluation. The tool just gives you a better memory.
Okay, let's scale this up. We've theoretically improved the local debrief and we've established that the instructor remains the absolute authority,
But if I am a program manager running a massive operation, maybe a statewide EMS training program across six different facilities, how does capturing local individual encounters help me build a better overarching curriculum.
To understand the scale, we have to examine white papers 04 and 05 which introduce the mechanisms that connect these isolated labs.
Okay. What are they called?
They call these concepts backall and the product space.
Backhole. That sounds like a logistics term like you're moving freight or something.
Well, the logistics metaphor is actually very accurate. Backhall is the governed mechanism that transports, normalizes, and preserves these local isolated encounters into a shared secure memory.
H how does that work in practice?
Imagine an instructor in facility A discovers a recurring subtle failure mode during a complex trauma scenario.
without backhole, that brilliant insight stays locked inside that one instructor's head or maybe it makes it into an email that gets ignored.
With backhole, that messy, chaotic encounter is transported and normalized, meaning the raw audio and timestamps are structured into a standardized timeline and it enters the network.
So, it's like taking the raw unrefined ore mined from a local simulation lab, loading it onto a truck, which is the back hole, and taking it to the refinery.
And that refinery is the product space.
That is precisely how it functions. The product space is where that compiled normalized experience is actively synthesized into actionable intelligence.
So, this is where the learn and improve stages of that loop really happen at scale.
Exactly. The system starts mapping out failure models across hundreds of encounters. It identifies reliable recovery pathways. It begins to analyze highly specific timing distributions.
Give an example of a timing distribution. Like what does that actually look like?
It might analyze 300 instances of a specific airway compromise scenario and determine exactly how many seconds on average a novice team has to correct a misintubation before The cascading physiological effects on the mannequin become unreoverable.
It takes isolated anecdotal incidents and turns them into structural measurable knowledge for your entire program.
Building a shared hive mind for EMS trainers is an incredible concept. But wait a minute. The moment you start talking about taking data from my local facility and sending it to a central refinery, my immediate concern is privacy and ownership.
Of course, that's a massive concern.
Are we just vacuuming up highly sensitive student data? their screw-ups, their hesitations, their worst moments, and just handing it over so Veimemed can train their massive commercial AI models.
This is a critical tension, and the white papers actually spend significant time addressing it. Vecommed uses a very specific strategic phrase. Every encounter becomes an asset.
Okay, but we have to look at how they define the word asset in their operating language because I imagine asset doesn't mean something VECD can just sell.
Correct. In their framework, ass set exclusively means retained learning value.
Retained learning value.
It absolutely does not imply that Vecommed assumes ownership of a person's biometric data or their performance history. The capture process is highly selective, right?
And the rights, retention policies, and any secondary uses of that data are strictly governed by your institution.
So, a recording of a student is not automatically an asset.
No, it is just raw data. The asset only emerges when those lawful governed observations are converted into trustworthy knowledge that helps you conduct a better debrief tomorrow.
Okay, that makes me feel better. The institution owns the data. Vecomet is just providing the architecture to refine it.
And let's bring this down to the actual reality of implementing this technology because this all sounds intellectually fascinating in a white paper, but as a training program manager, I can't just pause my entire curriculum, flip a switch, and hope this new software ecosystem doesn't completely derail my students.
No, you definitely can't.
So, How do you actually test something like this in a real functioning lab?
VEXMED's approach to implementation is deeply methodical. White paper08 outlines what they call a design partnership.
A design partnership.
They make a point to differentiate this from a generic software trial where you just, you know, install an app and see if people like the interface. A design partnership is an evidence-making relationship.
So what does this all mean? You are basically running a bounded controlled scientific experiment inside your own lab to prove whether the tech actually works for your specific use case.
Exactly. And they structure this experiment through four very specific evidence gates. Protocol, readiness, evidence, and decision.
Let's walk through those gates. How does gate one protocol work in practice?
In gate one, you define the boundaries. You absolutely do not roll this out to the entire facility,
right? Too risky.
You select one specific highly bounded workflow. Let's say you choose your pediatric cardiac arrest. scenarios.
You then establish a rigorous baseline before the software is ever turned on. You literally measure how long your current debriefs take, how much time the instructor spends taking notes, and you attempt to quantify what nuances are consistently lost for memory.
So, you know exactly what your normal looks like.
Then you move to gate two, readiness. I assume this is just making sure the tech actually turns on.
It is about freezing the product version so the software isn't changing underneath you while you test it. and ensuring the technical support model is fully integrated with your IT infrastructure.
That's smart. Once that is locked, you open gate three, evidence,
This is where we run the pediatric code scenarios with the system running in the background. What specific evidence are we looking for?
You are measuring against the baseline you set in gate one.
Was the system able to accurately capture the timeline despite the screaming alarms and overlapping voices? Most importantly, did it actually reduce the cognitive burden on the instructor or did it just give them another screen to stare at?
Yeah. Did it make the debrief faster and more targeted?
Exactly. Which leads directly to gate four decision. Based on that hard evidence, you decide whether to expand the system to other scenarios, modify the setup or completely pull the plug.
But I have to push back here based on my experience with software pilots. The vendor always always wants the pilot to succeed. They will bend over backward to massage the data to make it look like a win. That's the industry standard unfortunately.
Right. So what happens if the trial is just a disaster? What if the tech is too burdensome for the instructors and it completely distracts the learners?
This is perhaps the most revealing detail in all of the source documents. Vector Med explicitly dictates that quote stop conditions are a feature.
Wait, really? A negative result is considered a success.
A negative result is considered a valid accepted outcome of the scientific experiment. If the system creates an manageable review burden or if it negatively impacts the psychology of the learners in the room, the experiment is designed to stop.
Their ultimate goal is to rigorously empirically prove utility in the training environment before there is any conditional expansion into live field use.
So if it fails in your lab, they do not want you to push through the pain. They want the deployment to end.
That level of discipline is incredibly rare. And what is fascinating is how that discipline that deep respect for the gravity and reality of the medical profession is mirrored perfectly in their visual identity.
Oh, the brand guidelines. Yes.
Yeah. When you dive into the brand guide. PDF and white paper 01, they are meticulously controlling how they present themselves emotionally to you, the end user.
The entire brand identity orbits around a single core emotion. Earned competence.
Earned competence. It is a brilliant phrase. And when you look at the actual visual guidelines, the contrast with standard Silicon Valley aesthetics is uh honestly jarring.
It really is.
There are no glowing blue artificial intelligence brains. There are no floating sci-fi holograms or sleek futuristic heads-up displays.
They mandate a color palette built on deep greens, specifically hex code 2121 along with muted golds, stark blacks, and grays.
And they use a bold, heavily grounded typography called drummer.
And the imagery they permit is strictly focused on realistic medical training. Close-ups of gloved hands, CPR mannequins, the focused eyes of an instructor.
The voice and tone guidelines mandate that all communication must be calm, observant, and entirely serious. Looking at the brand guide, it honestly feels less like a software company and more like a military qualification patch or a specialized flight badge.
It looks like something a medical practitioner would be proud to earn through sweat and study. But I look at that stark seriousness. And I wonder, is it too severe?
How do you mean?
Well, we live in a world of consumer apps that use bright colors, gamification, and cheerful little pinging sounds to keep users engaged. Why lean so aggressively into this severe, almost austere aesthetic?
Because they intimately understand the psychology of their audience. The starkness is a highly calculated choice.
The core identity of the brand is the professional who can step into a chaotic terrifying room and confidently say, "When people are in danger, I know exactly what to do."
Preparing someone to say that is not a gamified experience. It does not deserve a cheerful pinging sound.
Uh Veco understands that highly trained professionals like the training managers listening to this deep dive are naturally rightly skeptical of any technology that promises to magically solve their operational problems.
Oh, absolutely.
By avoiding hype terms like revolutionary AI, rejecting futuristic imagery and embracing the dirt and sweat of the training room. They are actively building trust.
They are proving visually and semantically that they respect the life and death reality of the field you are ultimately preparing your learners for.
Exactly. They aren't treating your profession like a game because they know it's not a game.
So to synthesize all of this for you, the person actually running these training programs, Veco isn't trying to sell you a shiny AI that promises to replace human judgment. They are offering a highly disciplined training first learning loop.
By quietly capturing the messy complex transitions of learning by absolutely protecting your final authority as the educator and by compiling structural failure and recovery models through their back hole network. Their entire mission is to help your lab produce better medics.
And if we connect this all back to where we began, that brilliant messy ephemeral learning happening in your simulation lab right now, it leaves us with a trul fascinating possibility for the future of education.
What does that future look like?
Think about the sheer volume of data this system could compile over several years.
if this network successfully maps the exact timing, the specific physical actions, and the precise state transitions of how a novice eventually learns to recover from a critical error, how might that fundamentally change the way you design your scenarios tomorrow?
Instead of just running standard linear medical protocols, could you use this compiled int igence to intentionally reverse engineer the perfect mistake.
The perfect mistake,
right? Could you design a scenario perfectly calibrated to trigger a specific predictable cognitive overload in a student just so you can deliberately guide them through the recovery process, dramatically accelerating how fast they achieve true mastery?
Reverse engineering the perfect mistake that entirely changes the paradigm of how a curriculum is built. It takes that jagged, confusing evaporating reality of human learning and finally gives you the tools to read it, structure it, and scale it.
It really does.
Thank you for joining us on this deep dive into Veco. As you head back into the observation booth today, take a hard look at your next simulation scenario and just ask yourself, how could this fleeting encounter become a compounding asset for your entire program?