Ari Gesher, Matt Gordon and Julia Chmyz work at Kairos Aerospace, a Bay-Area-based company specializing in aerospace solutions for environmental surveying and digital mapping. Ari, Matt and Julia were interviewed in person during the 2018 Mapbox Locate Conference in San Francisco.
Describe Kairos Aerospace.
Ari: Kairos applies the notions of faster, cheaper, iterative cycles of technology to Aerospace. Specifically, with the mission of building sensors to spot very large leaks of Methane.
Julia: A less high-level description of Kairos — Kairos deploys aerial sensors, spectrometers, optical cameras, and thermal cameras to conduct large-scale surveys of assets from oil and gas companies, to survey those assets to discover things about them.
Matt: Kairos is a bunch of physicists and engineers who care about health and safety and climate change. We fly sensors and sell data about environmental pollutants (specifically methane) to oil and gas producers.
What led you each to Kairos?
Ari: I ended up at Kairos because the two original founders, Steve Deiker and Brian Jones, both worked at Lockheed for a long time, and they decided to start their own company. Steve’s wife worked with me at Palantir, and they knew that everything they were going to do was going to require a lot of heavy data processing, and that was not an area of expertise for them. They approached me for advice around what it would take to build a team with that kind of ability. That was late 2014. I was instantly interested, it sounded really, really cool… But, for reasons of childbirth, I was not about to switch jobs; I ended up being the original angel investor. Two years later I came on board as the director of software engineering.
Julia: Brian’s wife worked with the woman who was married to my grandfather. And so, my grandfather was actually another one of those original investors — This was 2015 — and he was saying to me, “Julia, there’s this great new company.” And I’m like, “Okay, Grandpa… I’m sure. That’s cool.”
Grandpa says, “They’re so great! They’re so great! You gotta send ‘em your resumé.” I was in school at the time (I’m a year out of college now), and I said, “Okay, fine grandpa, I’ll send ‘em my resumé.”
I hadn’t really looked into it, I just didn’t really want to work at this company my grandpa thought was so cool. But I sent my resumé, and I was really clear about this, I was like, “My grandpa’s really excited about this, but I’m not sure it’s such a good fit.” — expecting to give them an easy way out.
And instead, they wrote back and said, “We’re really interested! Your resumé looks great, we’d really love to have you on board.” So I came in and talked, and actually got to see for myself. And I was like, this looks really great. So I was an intern in the summer of 2016, when we were a third of size we are now. And then I came back full-time a year ago.
Matt: There’s a lot of funny history between Ari and I, which I won’t go into. I had just done my postdoc at Stanford in physics, and Ari recruited me to go work at Palantir. Then, about six years later, I quit and I was bumming around a bit, and making fire art.
Matt: Making fire art… yeah… and I thought I would go get a real job. Ari, at that point, was an angel investor, and he tried to recruit me into his current job.
Ari: That’s right, I tried to hire Matt for my current job.
Matt: And I turned him down to go start my own company, to develop online treatment for substance use disorders. Which, let’s say, the world was not ready for… [Polite chuckles] Mark my words: you’re going to see it.
And then about a year after doing that, Ari saw I was on the job market again, and asked me to come work at Kairos, on a team of four people – two full-times, an intern, and a couple of physicists who commited code to our code base (for better or for worse).
How many people are there now?
So it’s grown quite a bit?
Matt: Yeah. It’s moving.
Ari: Yeah there was sort of two different phases. The first two years, Brian and Steve quit their jobs and were literally in their garage in Los Altos, developing the hardware that is the heart of the methane sensor (which is the imaging spectrometer). And there’s pictures; like, one of them’s across the street, positioning a methane cell in the light path of a heliostat, the other one’s at the laptop with the original Mark-1 Spectrometer, making sure it worked.
Do they still have that?
Ari: They do — it sits on a shelf, and looks like a broken projector or something. [chuckles] So, the first two years was just validating that the hardware would work, and at the end of that, they had the design for what is today our production spectrometer, and the first production-designed unit (although we’re probably going to throw that one out pretty soon.)
The next two years have been developing both the operational side (How do we hook this thing up to a computer, and fly it, and collect data?), and also the software pipelines that sit behind it (How do we take that data off the instrument once it’s done? How do we upload it to the cloud, and develop the algorithms, from scratch, that turn that spectrographic data into the plume images that we have?).
Walk me through the process of: going out and sensing the area, to: you have a final product; and what that final product looks like.
Ari: The way that this works is that we’re given an area, a spot on the ground — the job we’re working on now is about 1,300 square miles?
Matt: We’re given a shapefile.
Ari: Right, we’re given a shapefile, and if we’re lucky, we’re also given a list of assets (another shapefile that tells us where all their wells and storage tanks and things are, so we can identify things once we find a plume over them). We then draw up flight plans to go fly over that area… like, if you look at it, you see the plane going back and forth like a lawn mower. And then, that data goes through the processing pipeline.
What comes out the other end are a stack of rasters that show us various measures of what the spectrometer has picked up. At a very rough level, what we’re actually sensing is a methane anomaly. Methane is everywhere in the atmosphere at some level; so it’s not “Is there methane here or is there no methane?”, but “Is there elevated methane?”
We use the large survey area, or chunks of it, to develop what we think the background levels of methane are in that area of the atmosphere. And then, we look for places in the data where there are elevated levels, and use that to interpolate a plume shape.
Example of a plume
One of the things we like to do at GeoHipster is geek out about the tools that people use; tell me about your day-to-day.
Ari: We’re mostly a Python shop. Very large amounts of effort dedicated to making GDAL install and compile correctly.
Matt: I do a lot of the GIS stuff at Kairos. There’s all the code for taking remote sensing data and GPS, and figuring out where that was placed on the ground. Then, taking all of that and creating GeoTIFFs out of that, with all the different metrics that we’re interested in.
Ari: And that’s all custom software, we don’t even use GDAL very much. We use GDAL to open the dataset that we write, but how we figure out what goes into each pixel is all ours.
Matt: Yeah, the ground placement of remote-sensed data is an art form… it’s interesting how much we’ve built from scratch. I think people with a lot of background in this probably know a lot of tricks and tools (and I’ve heard tell that there’s a book, but I’ve been unable to find it).
In terms of GIS nerdery: we used to do a lot of ad-hoc analysis in QGIS, and as we were increasing the number of reports we wanted to produce for customers, we wrote a QGIS plugin. It’s custom, and it’s not published anywhere because it’s specific to our workflow and our data, and it gives people summary information.
Anyone who has used QGIS will know that it’s like, incredibly powerful and can be incredibly frustrating. And if anyone from QGIS is reading this, I want them to know that I really appreciate the tool. We love it, and we would use something else if we thought it was better, and we don’t. There’s nothing else better.
Julia, you work on the tools that pilots use when they’re out collecting data. Can you tell us a bit about those?
Julia: There’s the feed that the flight operator sees in the plane, and the spectrometer frames that are being taken. There’s also all the IMU data that’s used for path stuff and all the later calculations… and this is our flight monitoring Mapbox Leaflet. The back end is built in Python, and the front end is in React.
Matt: Ari’s contribution was the X-Wing fighter.
Julia: The point of this is to make everything work as smoothly as possible — so the flight operators don’t have to spend their time staring at multiple log files, which is what they were doing before this.
Matt: So imagine a terminal, and just watching lines of term logs scroll past… in an airplane. In a very small plane.
Ari: Well, now that they use this, they say that they get kind of bored on the plane, because it gives them everything they need. In fact, we built this this tool not just to spit the information to the operator, but it also ingests all the raw data coming off the instrument; and we have a bunch of agents that watch that data for different conditions, and control the instruments.
It’s called R2CH4 as an homage to R2D2, who’s an astromech repair droid — and its primary job is not to save the universe, its primary job is just to make the X-Wing go.
I wouldn’t have caught that reference.
Well, CH4 is Methane sooooo… [makes the “ba-dum-tssssss” joke sound]
What do you do when you’re not at work – any hobbies? Matt, I heard about yours a little already: I know you’re a fire artist and you hang-glide?
Matt: I don’t hang-glide anymore, but yeah, I build weird Burner kinetic fire art. I’m making a fire Skee-Ball machine right now, where the balls are on fire. You get to wear big, fireproof kevlar gloves. I was going to bring it to Precompression, which is the pre-Burning Man party they do in SF, but the SF fire department nixed it.
Ari: I dabble in home automation. That’s kind of my tinkering hobby currently. I mean, I’ve had really good hobbies, but now my hobbies are basically my two children. But, you know… I used to be a DJ for a little while. I swear I used to have better hobbies — but I’ve really just been well-employed for like twelve years.
Julia: I spend most of my free time either outside, like hiking, or reading — real books with paper.
Ari: I thought that was illegal now?
Julia: It is here.
Just one last question for you.
Ari: 4-3-2-6! I’m glad you asked — it’s my favorite coordinate system.
Matt: 3-8-5-7 is way better, man.
Are you a geohipster? Why or why not?
Ari: Oh, absolutely. It’s interesting that all of us came to Kairos, not completely illiterate in the ways of GIS, but certainly not as well-steeped. And I was actually thinking about this on the way home: we have non-GIS operational data about what we do, but the core of what we do — everything is geo data. Like, there’s no non-geo data. And, what we’re trying to build is: taking a novel stream of data about the earth, and then running it through very, very modern software pipelines, to automate its processing, it’s production, all of that, in a way that requires understanding the bleeding edge of technology and blending that with GIS. And that’s what we spend all day doing.
Matt: I am geohipster because I make artisanal Geo data. And I’m opinionated about it. And I’m obnoxious. So, here a thing that I do, which is super geohipster: We produce a lot of stuff internally at the company, in WGS84 — which is not a projected coordinate system. It’s a geo-coordinate system — and I constantly complain about this. That we are producing GeoTIFFs in 4326, but we should be producing them in a projected coordinate system.
Julia: …And I want to tell you, we were doing all this way before it was cool.
Ari: One last thing — we use US-West 2 as our AWS data center, because it’s carbon-neutral (they run entirely on hydropower), so it fits in well with our overall mission.
Julia: I didn’t know that! I’m glad about that.
Ari: Suuuper hipster.
It is. Thank you guys!