On this episode of the NTP podcast we interview Aaron and Witold CTO & Project Manager at Cordel, an organisation that helps generate high detail point cloud maps of large scale infrastructure.
Hope you enjoy the interview!
Here you can source all the things we have talked about in the podcast whether that be books, events, meet-up groups and what’s new in the newcastle tech scene.
(00:20)
Tell us a bit about yourselves and your background in technology?
(03:30)
What are your roles within the Cordel?
(04:30)
What is Cordel technology and what are you currently working on with railway tech?
(10:58)
Who would be the perfect hire for Cordel?
(17:00)
Anything in particular that is publicly facing that you’re super proud of?
(31:30)
What are your thoughts on education and university as a pathway into tech?
(13:38)
How close does your team work with their customers?
(43:40)
What are your recommendations for self paced learning?
(46:00)
How do you manage to keep track of tasks?
(54:45)
How do people find you online if they would like to get in touch?
Thank you both for coming on i guess the easiest point to start give me a little bit of an intro on yourselves what you do all that good stuff uh we’ll start with aaron excellent so uh thanks i’ve been around in newcastle for uh longer than i probably care to say so been involved in technology well i finished uni kind of early 2000s and pretty much been doing startups ever since so first thing that i kind of started doing was mobile gaming so pretty much straight out of uni um won a kind of vodafone google games challenge in 2003 um and thought hey this is a hell of a thing to start doing so got a group of friends together kind of started doing that and then have really tried to avoid going to sydney over the course of the last 20 years so yeah the trick has always been trying to earn a sydney wage in newcastle which is uh a life goal shall we say um but yeah i’ve seen lots of people go down that path and just really keen to stay here and local so so yeah i’ve been working through you know then evolved into kind of mobile mapping because after phones got connectivity all of a sudden games weren’t really the best thing you could be doing with them uh then kind of a brief bit into social networking with makoki and then some local search with one three find and then around kind of yeah 2015 really uh met nick smith from airside and we started we’ve basically said hey let’s get a lidar sensor and how hard could it be to put a scanner together and that’s kind of where the kind of current iteration of cordell was really evolved from awesome but uh yeah i have much more colorful career and all over the place i’m a wood scientist from poland that somehow ended up in newcastle just pre-pandemic i’ve worked in construction i run my own business at some point i made a switch to i.t and i ended up working for a medical journaling company in sweden but the dark and the cold drove me out and i came to visit newcastle where i have some family and i absolutely fell in love decided to move and a few months later i ended up here i love that you still use the term wood scientists what’s the australian term for wood scientists probably a chippy
it’s great though because visuals actually done like after the pandemic when we’re moving back into the offices um we’re kind of looking at desks and rearranging everything and uh it’s all got um some very plain bits of wood and it’s all come up looking you know very custom very good for uh it’s a nice setup that we’ve got now so thank you yeah some carpentry skills come in handy when we need to redecorate office and things like that yes it’s good to have a diversity of skills because uh we don’t have things like a diversity of names in our company on the development team we have josh jordan jordan john so probably not our strongest sweden uh yeah and on the airside side we have jared is our surveyor and we’re just putting on a new surveyor also jared excellent excellent um i guess your positions within the company what do you do yeah so i’m chief technical officer and that is a fairly wide-ranging role with our company so and that’s everything from you know working with the hardware team to you know we’ve got our own scanning hardware uh right through to you know the automated analysis and then working with um be told on the ai team to actually drive you know a lot of the insights that people need from that kind of data so yeah and i am ai project manager this is very wide role as well i do a lot of ai training understanding of the data and the requirements of the customers a lot of technical understanding what it is that we’re actually trying to solve and what could we possibly solve with technology we have cool and sometimes you build desks that’s understandable yeah excellent it’s more of a hobby almost but you give him a job he’ll do it i guess broadly speaking uh what is cordell what is the company what is the aim that is an excellent question um so we’re kind of born and i think i just briefly touched on you know 2015 we’ve got a lidar sensor and we’re like this would be cool to kind of make a scanner and get it out there so um nick smith at the time was actually ceo of airsight and we kind of joined forces to start you know we initially wanted to build a lidar scanner for drones and that’s something that we actually have now it’s called nexcor it’s a product that we sell but through doing that this was fairly early on in the piece around 2016 we actually had sydney trains come to us and say well we actually want to do a drone lidar scan you know on this you know fairly big section the sydney trains network but um but there’s just a slight problem with that was all the bases so kingston smith airport which um you might know it it’s a small international job yeah so there’s basically no chance of ever flying a drone at night that’s like a 5k exclusion zone that you just can’t fly anything yeah so um so that was pretty much out because that was kind of most of the network so we’re like well what if we took this scanning rig that we’re designing for drones and just put it on a high rail or a train um and then because we’ve been coming from the drone space where weight is like such a key concern like every few hundred grams you add you know you’re cutting five ten minutes off your flight time like this is great we can throw all the sensors on here we’ll do cameras we’ll do you know a better gps and so so that was kind of the the concept for you know um the cordell side of things where we’re doing a lot of the the train based scanners um and cordell’s been kind of through a few iterations so initially it was corridor insights then it was smart corridor and then it was just corridor and now it’s cordell so um trademarking especially yes is a wonderful process that i uh hitting anyone who actually has to go through it but uh that’s the reason why it’s been through so many refinements sure and and things like that but um but yeah so now a lot of our work is actually you know we do build and sell the next core um drone scanners um and we have the same development team works across them and that shares a lot of technology with the train based scanners that we’ve actually got out there um but a lot of the work that we’re doing now is on the train based scanners and it’s effectively a mobile laser scanner which is not not really a new area but what’s new in what we’re doing is it’s relatively low costs we’re taking the lidar scanners that are coming out of autonomous driving and they’re kind of an order of magnitude cheaper than a lot of the kind of conventional kind of survey grade lidars that have been around for a bit longer um so we can use those they’re weatherproof they can be permanently mounted that kind of thing because autonomous driving is you know it’s effectively an always-on type scanner they’ve been designed for that kind of use case so we’re trying to get you know that that cost of capture of all that data down by an order of magnitude sure and reasonably successfully too which then creates the next thing which is a data problem and a lot of networks already have this because they’re getting linear scans done and they get you know effectively a point cloud at the end of it which you can make measurements in and that’s well and good but then you actually need an engineer to sit down and make all the measurements that you need and it can actually be just as time consuming as going out and manually doing it so it’s uh yeah that’s where we’re kind of starting to add a lot of value into there so sure yeah one thing we’re also focusing on as opposed to specialized scanning when you need a specialized rig that just goes very slowly and the rail or whatever piece of truck needs to be blocked off we’re focusing on having our equipment on trains that are in service or specialized trains that travel the network anyway yeah and so here in australia we’ve working with rtc so they manage a good chunk of our railway here in australia especially here in new south wales and they’ve actually got um a track geometry cart the ak car okay travels all around australia and their primary mission really is measuring the track geometry so you know where any defects in the rails and things like that so but we’re actually got our sensor suite on there as well cool and we’re effectively just along for the ride so we’ve got a whole thing set up um it’s automated it’s just capturing whenever it’s moving um and we’ve got 4g telemetry in so we can see up on our endeavor and wall we can see you know health indicators green is good red’s bad yep don’t see much red thankfully but uh but yeah so we can monitor it pretty much anywhere where there’s a phone signal or not yeah cool and it’s actually a bit of a testament to some of our mobile phone networks because it stretches a lot further than i would have thought it goes in pretty remote areas so and to put that in scale we scanned up about 60 000 kilometers something like 40 000 kilometers last year which is about circumference of earth i think yeah yeah so approximately that so yeah about 60 000 k’s a year we’re kind of doing just on that one scanner yeah yeah but yeah as i said that creates a big data problem yeah so you solve one problem the making it cheap to capture and then you know you’ve effectively created you know an even worse data problem that you have yeah yeah yeah um what i guess what is the tech stack of core dell um i imagine there’s going to be a lot of effort uh required in basically storing and making that data useful um i imagine there’s a lot of effort required in physically just developing visualization sporting that sort of thing uh i guess the perfect hire for cordell what skills do they have yeah and this is something that you know it varies obviously by um you know piece that we’re working on but uh we do use dotnet fairly extensively because like starting at you know the hardware level we have um effectively a rover that’s doing all the data logging for the mobile laser scanning so it’s like pulling the lidar information in it’s then timestamping the camera feed to the gps the same as the lidar keeping the gps and you know kind of really collating all that data um so that’s all you done in net we actually have like a linux environment on a single board computer um and it used to be mono but now we can run.net core you know even better so uh so we run that kind of at the coal face essentially like as we’re capturing uh and then we have you know uh normally something watching that there that is effectively just a gui that’s on a screen somewhere so the operators can can actually see whether or not they actually look at it is another thing but it’s there but it’s there so and then we pipe uh telemetry up to our web service um and we’re using.net on the website as well so we’re kind of using nbc.net and uh and i’m not big on the website of things but we’re bringing in a bunch of new javascript frameworks and bits and pieces so so then on the website yeah we’ve got you know quite a lot of technology going on there because all this data we’ve got to store it somewhere so we’re putting it into s3 containers in aws or azure storage blobs um and so you know that cloud back-end and infrastructure is kind of a fairly key key component for us absolutely but the dot-net side you know that’s where we’re accessing it and that kind of thing but there is also you know after that dot has been uploaded you need to be able to view it and this is you know something that we’ve found is a a big problem for a lot of our clients is oh yeah we’ve got to light our scan done it’s on you know john’s desk if you need it you know go tap him on the shoulder and put the drive in your machine and then open up an engineering piece of software to get the bitter dot whereas we want to make it all kind of web accessible so that you know and so we have a lot that goes into the visualization of that data so that you know effectively they can click on a map and go hey you know something happened on this area of track they can review the video footage look at the lidar point cloud um any of the automated analytics for you know that stretch um and so on that side there’s kind of a lot of front end work too so viewing the point tree uh the point cloud we use poetry and there’s a lot of javascript to you know run just to run basically the web app really well on the uh on the client side but then on the server side you know there’s automated processing frameworks that we have to do and we have you know about 12 stages that data’s going through so um so there’s a lot of kind of heavy lifting with dotnet and utilizing a lot of kind of the 3d side of things on the back end so in that area you know the map becomes a lot more intense and we’ve got uh actually well we have some people with point cloud experience and also some ex-gaming guys as well because it turns out a lot of the 3d math translates very well from you know the gaming space into um so that’s kind of you know a lot of net there but it’s used in very different ways on you know things like the rover going around the website and the back end um after we’ve kind of one of the steps in our automated pipeline we actually do like a lot of ai analytics on it and that’s where kind of vital and the ai team basically gets sent a bunch of data and this is a fairly loosely coupled system and we use a a slightly different technology stack on that side yeah we use mxnet and python on the backend for machine learning and we manage all our jobs and models through a web app we built and that runs on meteor which is a framework based on node and react combined with that that’s fairly simple and then we communicate with the rest of our own system through an api cool so we we have our system dockerized and ready to like we run it locally on our local machines but if we need to deploy it at scale we have everything dockerized so ready to scale up to azure if we need to or aws yeah which is always interesting a lot of those instances of some of them more expensive than employees but
um but it’s worked really well having that slightly different stack for you know kind of everything and then the ai is really kind of its own beast um especially uh there are many challenges in making a production system using python i’ll say but uh but it is also the best language to be doing a lot of ai and machine learning in because anytime a new paper comes out they’re not using c sharp they’re not using.net it’s python is the example and if you want any hope of even replicating what they’ve done then it it really needs to be we’re looking to slowly migrate into julia but that’s a long-term project so yeah yeah yeah and that’s worked you know reasonably well for us th