Launch HN: Enhanced Radar (YC W25) – A safety net for air traffic control

166 points by kristian1109 ↗ HN
Hey HN, we’re Eric and Kristian of Enhanced Radar. We’re working on making air travel safer by augmenting control services in our stressed airspace system.

Recent weeks have put aviation safety on everyone’s mind, but we’ve been thinking about this problem for years. Both of us are pilots — we have 2,500 hours of flight time between us. Eric flew professionally and holds a Gulfstream 280 type rating and both FAA and EASA certificates. Kristian flies recreationally, and before this worked on edge computer vision for satellites.

We know from our flying experience that air traffic management is imperfect (every pilot can tell stories of that one time…), so this felt like an obvious problem to work on.

Most accidents are the result of an overdetermined “accident chain” (https://code7700.com/accident_investigation.htm). The popular analogy here is the swiss cheese model, where holes in every slice line up perfectly to cause an accident. Often, at least one link in that chain is human error.

We’ll avoid dissecting this year’s tragedies and take a close call from last April at DCA as an example:

The tower cleared JetBlue 1554 to depart on Runway 04, but simultaneously a ground controller on a different frequency cleared a Southwest jet to cross that same runway, putting them on a collision course. Controllers noticed the conflict unfolding and jumped in to yell at both aircraft to stop, avoiding a collision with about 8 seconds to spare (https://www.youtube.com/watch?v=yooJmu30DxY).

Importantly, the error that caused this incident occurred approximately 23 seconds before the conflict became obvious. In this scenario, a good solution would be a system that understands when an aircraft has been cleared to depart from a runway, and then makes sure no aircraft are cleared to cross (or are in fact crossing) that runway until the departing aircraft is wheels-up. And so on.

To do this, we’ve developed Yeager, an ensemble of models including state of the art speech-to-text that can understand ATC audio. It’s trained on a large amount of our own labeled ATC audio collected from our VHF receivers located at airports around the US. We improve performance by injecting context such as airport layout details, nearby/relevant navaids, and information on all relevant aircraft captured via ADS-B.

Our product piggy-backs on the raw signal in the air (VHF radio from towers to pilots) by having our own antennas, radios, and software installed at the airport. This system is completely parallel to existing infrastructure, requires zero permission, and zero integration. It’s an extra safety net over existing systems (no replacement required). All the data we need is open-source and unencrypted.

Building models for processing ATC speech is our first step toward building a safety net that detects human error (by both pilots and ATC). The latest system transcribes the VHF control audio at about ~1.1% WER (Word Error Rate), down from a previous record of ~9%. We’re using these transcripts with NLP and ADS-B (the system that tracks aircraft positions in real time) for readback detection (ensuring pilots correctly repeat ATC instructions) and command compliance.

There are different views about the future of ATC. Our product is naturally based on our own convictions and experience in the field. For example, it’s sometimes said that voice comms are going away — we think they aren’t (https://www.ericbutton.co/p/speech). People also point out that airplanes are going to fly themselves — in fact they already do. But passenger airlines, for example, will keep a pilot onboard (or on the ground) with ultimate control, for a long time from now; the economics and politics and mind-boggling safet...

117 comments

[ 3.3 ms ] story [ 209 ms ] thread
Love this. I sent this to a few of my military and commercial pilot friends to see if they are interested in talking / helping out. Good luck solving this problem!
Appreciate your help! Always keen to discuss with aviation folks
laudable. I'm hoping the radar UX will move away from 1930s 2d oscilloscope sweeper one day, instead of AI alucinating speech.
Yes, this speech technology is day one, but we've been talking a lot about what a future tower should look like. Augmented radar is complex and very interesting.

Little known fact: some towers don't even have radar at all. They're just up in the cab with binoculars.

I've watched a lot of Aircraft investigation stories, with incidents like this happening, sadly, rarely there are people who are interested in an intersection of the both technologies to be able to find a proper functioning solution and I think this is pretty interesting stuff you guys are doing, if you'd been working on re-inventing the wheel with new software to automate flight trajectory management, I'd not be as amazed, I think you guys have really taken out the time to understand the problem and worked on a potential solution, that could have a major impact.

> This system is completely parallel to existing infrastructure, requires zero permission, and zero integration. It’s an extra safety net over existing systems (no replacement required). All the data we need is open-source and unencrypted.

The part where you explain that it is integrable in the existing chain of command at Airports is proof enough.

Wishing you all the best for your venture.

Thank you — yes, it's super important to us to find the wedge where we can actually ship quickly. Nothing beats feedback from real products in the real world. As much experience as we have being pilots, there's so much to learn on the control side.
This sounds awesome. As a GA pilot and software developer, I can tell you that voice comms and the need to work with them will be around for a long time. Just look at all the feet dragging just to get ADSB adopted. I'd love to find out more and possibly get involved. Are you looking for help?
Appreciate it; we're not quite hiring, but I'll come back to this when we start.
Any plans on open-sourcing your ATC speech models? I've long wanted a system to take ATIS broadcasts and do a transcription to get sort of an advisory D-ATIS since that system is only available at big commercial airports. (And apparently according to my very busy local tower, nearly impossible to get FAA to give to you).

Existing models I've tried just do a really terrible job at it.

I've thought about the same thing; transparently, we were trying to get a reliable source of ATIS to inject into our model context and had the same issue with D-ATIS. What airport are you at? Maybe we whip up a little ATIS page as a tool for GA folks.
That would be awesome! My airport is KPDK (sadly it doesn't have a good liveatc stream for its ATIS frequency).

I did collect a bunch of ATIS recordings and hand-transcribed ground-truth data for it a while ago. I can put it up if that might be handy for y'all.

If you're willing, that'd be great. I think our model will do well out of the box, but more data is more better as they say.

I spent a lot of time out at PDK when I worked briefly in aircraft sales. Nice airport!

Let me work on this and come back! I think we can ship you an API for ATIS there...

Hey this seems like a great idea and has a lot of promise! If you're looking for a frontend / UI / UX engineer, I'd love to get involved. Either way, best of luck! Let me know - hello (at) joshmleslie.com
Why the focus on speech-to-text and not purely on radar and predicting 3D movement or at least shutting down in 3D space a runway that is occupied?
Great question - the fundamental entry point of the air traffic system is this VHF control audio. Everything is downstream of that. Trajectory planning misses the intention signal on the frontend.

The example in the original post is actually a good case study for why trajectory planning alone breaks down. By the time the aircraft are on a predictable collision course with each other, you've lost 10+ seconds of potential remediation time that you would've had if you detected the error when it was spoken. Those 10 seconds really matter in our airspace.

One major issue is that radar doesn’t work very well at low altitudes. The lower you aim the beam, the more ground clutter you pick up (trees, buildings, clouds, etc).
Yes, this is a challenging issue in ATC. Also, some safety systems are turned off or diminished at critical phases of flight (near the ground) because of the noise problem (TCAS, for example).

ADS-B helps with this as it's self-reporting. And systems like ASDE-X are useful to track objects once they hit the ground. But low-altitude deconfliction is a big problem.

My understanding is that TCAS is disabled not due to radar limitations, but because it isn't altitude-aware and will happily generate an avoidance command that results in CFIT.
The systems I've flown have definitely been altitude aware. They won't alert if the plane is sufficiently deconflicted above or below.

One problem for sure is that when you're close to the ground you have to be careful about buildings, cell towers, etc. Terrain is one thing, but when you're a few hundred feet AGL, you could quickly be in the way of tall structures if a TCAS alert goes off.

It's been a long time since I worked in flight simulation (full flight simulators and simpler pilot training devices, including simulating TCAS), but I believe at that time TCAS would be switched to a mode in which it only alerts of "Traffic" instead of providing avoidance instructions precisely _when_ entering busier airspace -- e.g. airport proximity. In that environment it was undesirable for TCAS to be giving instructions. That seems like the environment in which Enhanced Radar's (future) product(s) could be of most interest.

(By the way, I believe EGPWS would take priority over TCAS anyway.)

I'm the developer of a speech-to-speech tool for tactical radar control for a combat flight simulator (https://github.com/dharmab/skyeye). My users have often asked to expand it to ATC as well, usually under the impression that it could be done trivially with ChatGPT. I love that I can now link to your post to explain how difficult this problem is! :)
Being unfamiliar with DCS' architecture, I expected this repo to be in Lua or something. I was surprised to find a very polished, neatly structured, well-documented Go service, haha. Very cool!
I just want to say this is such a pivotal problem to solve in this era of aviation. Looking forward to the progress you all make here. If you need any help on the product or sales side, let me know!
> The latest system transcribes the VHF control audio at about ~1.1% WER (Word Error Rate), down from a previous record of ~9%.

I'd be curious about what happens when the ASR fails. This is not the place to guess or AI-hallucinate. As a pilot, I can always ask "Say Again" over the radio if I didn't understand. ASR can't do that. Also, it would be pretty annoying if my readback was correct, but the system misunderstood either the ATC clearance or my readback and said NO.

Good and fair questions.

In the very short term, we're deploying this tech more in a post-operation/training role. Imagine being a student pilot, getting in from your solo cross country, and pulling up the debrief will all your comms laid out and transcribed. In this setting, it's helpful for the student to have immediate feedback such as "your readback here missed this detail...", etc. Controllers also have phraseology and QA reviews every 30 days where this is helpful. This will make human pilots and controllers better.

Next, we'll step up to active advisory (mapping to low assurance levels in the certification requirements). There's always a human in the loop that can respond to rare errors and override the system with their own judgement. We're designing with observability as a first-class consideration.

Looking out 5-10 years, it's conceivable that the error rates on a lot of these systems will be super-human (non-zero, but better than a human). It's also conceivable that you could actually respond "Say Again" to a speech-to-speech model that can correct and repeat any mistakes as they're happening.

Of course, that's a long ways from now. And there will always be a human in the loop to make a final judgement as needed.

> Looking out 5-10 years, it's conceivable that the error rates on a lot of these systems will be super-human (non-zero, but better than a human). It's also conceivable that you could actually respond "Say Again" to a speech-to-speech model that can correct and repeat any mistakes as they're happening.

This is effectively AGI.

And I've not seen anyone reputable suggest that our current LLM track will get us to that point. In fact there is no path to AGI. It requires another breakthrough in pure research in an environment where money is coming out of universities.

AGI is a moving target, but agreed, lot's more research to be done.
It isn't AGI, it is domain specific intelligence.
Thanks for that. It must be exciting to be applying software skills to aviation. Life goals!

To me, speech to text and back seems like an incremental solution, but the holy grail would be the ability to symbolically encode the meaning of the words and translate to and from that meaning. People' phraseology varies wildly (even though it often shouldn't). For example, if I'm requesting VFR flight following, I can do it many different ways, and give the information ATC needs in any order. A system that can convert my words to "NorCal Approach Skyhawk one two three sierra papa is a Cessna one seventy two slant golf, ten north-east of Stockton, four thousand three hundred climbing six thousand five hundred requesting flight following to Palo Alto at six thousand five hundred," is nice, but wouldn't it be amazing if it could translate that audio into structured data:

    {
    atc: NORCAL,
    requester: "N123SP",
    request: "VFR",
    type: CESSNA_172,
    equipment: [G],
    location: <approx. lat/lon>,
    altitude: 4300,
    cruise_altitude: 6500,
    destination: KPAO,
    }
...for ingestion into potentially other digital-only analysis systems. You could structure all sorts of routine and non-routine requests like this, and check them for completeness, use it for training later, and so on. Maybe one day, display it in real time on ATC's terminal and in the pilot's EFIS. With structured data, you could associate people's spoken tail numbers with info broadcast over ADS-B and match them up in real time, too. I don't know, maybe this already exists and I just re-invented something that's already 20 years old, no idea. IMO there's lots of innovation possible bringing VHF transmissions into the digital world!
Who gave you our event schema!? ;)

Kidding aside, yes, you're exactly right. We're already doing this to a large degree and getting better. Lots of our own data labeling and model training to make this good.

Best of luck to you. Finally a Launch HN that's important, potentially life-saving work.
One of the challenges I imagine you'll face as you move towards active advisory is that the more an alerting tool is relied upon, the more an absence of a flag from it is considered a positive signal that things are fine. "I didn't hear from Enhanced Radar, so we don't need to worry about ___" is a situation where a hallucinated silence of the alerting tool could contribute to danger, even if it's billed as an "extra" safety net.

I imagine that aviation regulatory bodies have high standards for this - a tool being fully additive to existing tools does not necessarily mean that it's cleared for use in a cockpit or in an ATC tower, right? Do you have thoughts about how you'll approach this? Also curious from a broader perspective - how do you sell any alerting tool into a niche that's highly conscious of distractions, and of not just false positive alerts but false negatives as well?

Yes, fair points. In talking to controllers, this has already come up. There are a few systems that do advisory alerting and controllers have expressed some frustration because each alert triggers a bunch of paperwork and they are not 100% relevant.

There are lots of small steps on this ladder.

The first is post-operational. You trigger an alert async and someone reviews it after the fact. Tools like this help bring awareness to hot spots or patterns of error that can be applied later in real time by the human controller.

A step up from that is real-time alerting, but not to the main station controller. There's always a manager in the tower that's looking over everyone's shoulder and triaging anything that comes up. That person is not as focused on any single area as the main controllers. There's precedence for tools surfacing alerts to the manager, and then they decide whether it's worth stepping in. This will probably be where our product sits for a while.

The bar to get in front of an active station controller is extremely high. But it's also not necessary for a safety net product like this to be helpful in real time.

I think it would be handy to have it as a check. If I get an alert about a potentially incorrect readback, then I can call back for clarification.
> I'd be curious about what happens when the ASR fails.

When, not if. The "artificial intelligence" as it is presently understood is statistical in nature. To rely on it for air traffic control seems quite irresponsible.

Great use of AI models, since language is likely enough to infer the situation.
Language is critical input data. Appreciate it!
I'm loving it. As a private pilot working (slowly) towards my CFI rating - this is also such an opportunity to integrate it into training devices.

Bulk of the instrument flight training is "mindgames" anyways - you see nothing other than instruments, your "seat in the pants" is likely to cheat you..

Possibly going a step further, the state of teaching aids available to CFIs is pretty sad, with online quizzes and pre-recorded videos being the pinnacle of what I have experienced... this would be an awesome opportunity to try and build "automatic CFI" - not counting for AATD time under current rules but better than chair-flying (the process of imagining a flight and one's reactions).

I have a bit of an ax to grind with the state of pilot training tools. There's some interesting new work being done here, but agreed lots to do in this realm. What does a compelling "automatic CFI" look like to you?
> Bulk of the instrument flight training is "mindgames" anyways - you see nothing other than instruments, your "seat in the pants" is likely to cheat you..

Eh, I guess I can flex a little. Living in the Pacific North West, I do not have to play mind games. I can almost get IMC delivered on demand. :P

The teaching tools nowadays available to CFIs are fantastic. Get yourself X-Plane 12, Honeycomb yoke, throttles and rudders and a decent aircraft like the Challenger 650 and you are good to go. All of this can be had for around 1000 bucks and is the best investment anyone can make to stay current. I used to be a CFII a long time ago (1995) when sims where expensive and only the best flight schools had them. Back then, instrument training was flying long and boring hours under the hood and if you were lucky, one or two hours actual IFR. Most people were afraid to fly real IFR after the rating, since they were aware that the training was so bad, myself included. A couple of hours flying a PC sim fixed this problem once and forever and made a huge difference in real world flying.
> To do this, we’ve developed Yeager, an ensemble of models including state of the art speech-to-text that can understand ATC audio.

Listening on multiple channels might help at busier airports. Ground, ramp, approach, departure, and enroute are all on different channels. Military aircraft have their own system. (That may have contributed to the DCA accident.)

Something like this was proposed in the late 1940s. The FAA was trying to figure out how to do air traffic control, and put out a request for proposals. General Railroad Signal responded.[1] They proposed to adapt railroad block signalling to the sky. Their scheme involved people listening to ATC communications and setting information about plane locations into an interlocking machine. The listeners were not the controllers; they just did data entry. The controllers then had a big board with lights showing which blocks of airspace were occupied, and could then give permission to aircraft to enter another block.

Then came radar for ATC, which was a much better idea.

[1] https://searchworks.stanford.edu/view/1308783

It's been interesting to see that a product as simple as combining data from multiple frequencies at once has been really compelling to folks. Can't tell you the number of times we've heard "wait, can you compile ground, tower, and approach in one place?"... "... yes, of course."

Military aircraft are typically equipped with UHF radios (in addition to civilian VHF). Many of the same systems apply, just a different RF band. And we're in the process of adding UHF capabilities to our product as a lot of these military aircraft land at civilian airports for training exercises.

I can't imagine what would've happened if we adopted block signaling for ATC ...

> I can't imagine what would've happened if we adopted block signaling for ATC ...

You don't have to imagine. We already do in many places. The North Atlantic Tracks are essentially exactly that. Aircraft give position reports and estimates, those positions reports are used to decide whether an aircraft can climb though which levels etc.

It's also used extensively in an IFR non-radar environments. Exactly why aircraft have to cancel IFR at uncontrolled airfields in the US or under a procedural ATC service in the UK. You hear it a lot around the Caribbean and Bahamas too.

I have two questions:

1. Good overview of the technologies you are using, but what product are you planning on building or have built? I understand what you are doing and it's "extra safety over existing systems" but how does it work for the end user? Is the end user an ATC or a pilot?

2. You will find that introducing new systems into this very conservative field is hard. I've built avionics and ATC electronics. The problem isn't normally technology, it's the paperwork. How do you plan on handling this?

1. Our first product is post-op review at airports. We're selling that to airport managers who use our system for training and incident review. Today, when a ground ops vehicle (for example) makes a mistake, the airport manager has to note the incident, call the tower, wait a week for them to burn a CD of the audio, scrub through to find the relevant comms, go to a separate source to pull the ADS-B track (if available), fuse all that together, and review with the offending employee. Our product just delivers all that data at their fingertips. For training, we also flag clips where the phraseology isn't quite right, etc. Obviously this isn't the long term product, but it gets us to revenue quickly and side-steps regulation for now.

2. Agree

[edit] (oops, sorry, seeing your edit)

2. The regulation allows for escalating assurance levels. We'll start with low assurance (advisory) and climb that ladder. We're definitely not naive about it; this will be hard and annoying. But it's inconceivable that someone won't do this in the next 10 years. Too important.

Thank you for the detailed reply. Your first product sounds like something that is needed. I wish your startup very good luck and will be watching your progress.

Do ground vehicles also have GPS trackers with a radio transmitter, or do they just use normal ADS-B?

I was thinking along the same exact lines:

Why do we still rely on analog narrowband AM voice over VHF to do vital comms like air traffic control? Same way as we did in the 1940s!

We should be transcribing the messages and relaying them digitally with signatures and receipt acknowledgement.

Isn't it because AM audio is still understandable under very suboptimal conditions where digital might not get through? Digital narrowband data modes tend to pass very small amounts of data
Quite the opposite. For short messages digital modes can employ layers of redundancy, auto carrier recovery, error correction at all layers all while yielding lower power requirements, and longer distance.
FAA actually tried moving to digital voice (has benefits wrt airband congestion) but it didn't go anywhere. I believe a lot came down to the minimal benefit over current solutions, plus the coordination and safety implications of actually making the switch. Tough for an FAA official to pull the trigger on a rollout that has even 0.1% chance of an aircraft crashing.
A lot of ATC seems to use lowest common denominator tech so that you an fly a Cessna into JFK.
But only after 1am when you're not fighting a 15 knot headwind with an A320 cleared number two.
AM modulation is perfectly justified in this context: if two (or more) stations accidentally transmit at the same time, this will be noticed. Using FM, only the stronger signal wins and the other signal remains undetected. The advantage over digital transmission is the lack of coding overhead - the voice reaches the receiver without any time delay.
The justification still holds, but better tech with the same benefits exists nowadays.

As far as digital decoding delay is concerned, this is a negligible number if implemented correctly.

Maybe the future is structured electronic messaging with the humans in the loop.

Like, check in with the controller but most messages are sent electronically and acknowledged manually.

I have your clearance, advise when ready to copy, then you write everything down on kneeboard with a pencil and then manually put it in the navigation system, is a little archaic.

certainly speech to text is a useful transition but in the long run the controller could click on an aircraft and issue the next clearance with a keyboard shortcut. then the pilot would get a visual and auditory alert in the cockpit and click to acknowledge.

I would hope someone at NASA or DARPA or somewhere is working on it. And then of course the system can detect conflicts, an aircraft not following the clearance etc.

This already exists and is used in much of the US and extensively in Europe for airlines. Look up Controller Pilot Data Link Communications (CPDLC).
> Maybe the future is structured electronic messaging with the humans in the loop.

There already is: Controller Pilot Data Link Communications (CPDLC).

Get an instruction, press to confirm.

At the moment, this is only used for certain types of things (clearances, frequency changes, speed assignments, etc.) along with voice ATC.

https://en.wikipedia.org/wiki/Controller%E2%80%93pilot_data_...

The problem with datalink systems is they are poor substitutes for immediate control & confirmation. My co-founder Eric wrote a short piece about this: https://www.ericbutton.co/p/speech. This is why they are mainly relegated to low-urgency en-route & clearance delivery.
interesting! have PP but haven't flown really last couple of decades.

I feel like, with proper UX in the cockpit and on the controller console, making it easy to send/acknowledge the clearance, and intrusively demanding immediate acknowledgment for important messages, with the controller able to talk to the pilot if it isn't immediately acknowledged, structured messages would save time, be more accurate, allow automated checks, i.e. be a superior substitute.

UX needs a ton of work and human factors validation, and would take 20 years to implement. But if you were starting from a blank slate it seems like the way to go!

He’s write about the bandwidth and latency of voice, but the problem is that you can’t immediately know who should react to instructions. “GO AROUND IMMEDIATE!” - now all the pilots on frequency are wondering who’s the addressee

Also, AM voice on VHF is not full duplex and the blocking problem is very real and could be addressed potentially

also requires fairly expensive equipment (FMS with FANS support)
Your system fuses ATC speech recognition, NLP, and ADS-B signals to detect and mitigate human error in air traffic control. Given the rapid advancements in multimodal AI, have you explored integrating visual data sources (e.g., satellite imagery, radar feeds, or airport surveillance cameras) to further improve situational awareness and error detection? What challenges do you foresee in making Yeager more contextually aware using additional modalities?
Yes, this is an excellent prompt and we're working on it. One problem is a lot of these visual sources require permission, integration, and regulation. That's going to move slower than something we can proceed directly with (VHF antennas).

I believe scaling laws will hold as we start to feed all of this context data into an integrated model. You could imagine a deep-q style reinforcement learning model that ingests layers of structured and visual data and outputs alerts and eventually commands. The main challenge I foresee here will be observability... it's easy enough to shove a ton of data into a black box and get a good answer 98% of the time. But regulation is likely to require such a system to be highly observable/explainable so the human can keep up with what's going on and step in as needed.

Looking further into the future, it's plausible the concrete structures of today with humans looking out windows will be replaced with sensor packages atop a long flagpole that stream high-res optical/ir camera data, surface radar, weather information, etc into a control room with VR layers that help controllers stay on top of busier and busier airspace.

The current rash of airline incidents reminds me of the assembly instruction: Torque fastener until you hear expensive sounds and then back off a quarter turn.

We've accelerated past our capabilities and need to slow down. ATC has incentive to slot takeoffs and landings as close as possible, but that is in tension with the goal of safety.

> Air traffic control from first principles looks significantly more automated.

We have a system 'designed' by history, not by intention. The ATC environment is implemented in the physical world, everyone has to work around physical limitations

Automation works best in controlled environments with limited scope. The more workaround you have to add, the noisier things get, and that's why we use humans to filter the noise by picking important things to say. Humans can physically experience changes in the environment, and our filters are built from our experiences in the physical world.

Anyway, sorry that isn't a question.

No stress, appreciate you chiming in.

> We've accelerated past our capabilities and need to slow down.

This is a super interesting meditation. As much work as there is to be done now, the demand for air traffic is growing and power laws are concentrating it into tight airspace bubbles. It would behoove us to figure out how to make airspace more dense without compromising safety. There's lots of good economic incentive for this.

> how to make airspace more dense without compromising safety.

My suggestion would be to make things as repeatable and consistent as possible. This would mean forcing some airports to change their practices to be consistent with everyone else, and forcing physical layout changes and construction. Unfortunately an app can't do that =( ... and the benefits are on the other side of a paradigm shift, so it's hard to make it happen naturally.

>> more dense

Large, high passenger capacity airliners have gone out of style but that would have been the best way to get passenger density up.

IMO one of the only interesting things "block chain" tech ever produced that had real world value and potential to save lives.

https://aviationsystems.arc.nasa.gov/publications/2019/SciTe...

Curious if OP has seen this paper / project before?

Interesting paper, thanks for sharing here. Encrypting ADS-B is a whole discussion... in my opinion, it's a good thing to have largely public location data for folks to consume and study.

What were your main takeaways from this paper?

What's your intended model for this? Pitch to air traffic management as their additional safety net or sell direct to operators/pilots as a warning service?
Short answer: both.

Pitching air traffic management is going to be a years long process.

Getting certified for on-board avionics is similarly challenging.

In the meantime, we'll get better and better at monitoring the airspace system and deploy that technology into unregulated applications like post-operational roles.

Pretty cool use of NLP. I wonder if training schools can use this to train their students to improve their readback detection. IMO, one of the hardest things as a pilot in training in the first few months is getting feedback on my communication with the tower.
Exactly, this is one of our early beachheads. If we can train better pilots, the airspace system gets safer immediately.
(comment deleted)