In my previous post about Unearthed Perth I explained why I thought it was so important to Perth and the Startup Community here. So I thought I’d let you know how it went from my point of view. There will be official pronouncements from the Unearthed lads, and this isn’t that… this is my experience of the event as a participant.
We meet up on Friday evening, and have some talks. The Rio video was interesting, but I suspect every single one of us was more impressed by the workstations with about a dozen monitors in a huge hemisphere than we were by any of the actual mining stuff. But it did drive home just how huge mining is in WA. For a lot of us who work in different sectors it can be overlooked; seeing the scale of everything mining-related does ram home the importance of mining here.
The next hurdle was that the data quality is a mess. We were looking at the Goldfields data, and they’ve obviously bought a commercial product then tried to get their operation to fit into it, and there’s some rough edges. We discover Truck 201 to be almost godlike in its ability to haul thousands of tonnes of ore in a single load, sometimes spending less than 10 seconds to do that, and never spending a single second queuing. We promptly rename our team TR201 in its honour (though Sam’s suggestion of “Truck Norris” came a close second). We learn later that this truck is the one they use to reconcile estimates and actuals in the system – they don’t just alter the values of ore stockpiles, they create a fictional cycle for TR201 to move the notional ore.
Talking to the other teams, we find the same issue for most of us: We don’t know what problems to solve, or what value there is in different parts of the system. Our team got a reasonably good handle on the ore-haulage cycle for Goldfields, so we start working on that, but we don’t really know where the value is or what the existing system provides. A couple of the other teams seemed to have good ideas about some problems, and a couple of teams tackled the known problems in interesting ways, but most of us were clueless and struggling to work out where the value is.
We had some help; Sean from TheMiningHub.com dropped in to share his insight, and we showered him with questions about how a mine actually works and what might be useful, but our original aim of modelling the ore-haulage cycle was foiled by data quality, lack of map co-ordinates, and misunderstanding the data. Another team, The Froys, did manage to do this and map it all out very impressively and went on to win the map prize (well done chaps, good job!). Because we couldn’t work out where the value was in the data, and the data wasn’t clear enough to enable simple modelling, all we could do was summarise it to a neat-looking dashboard and hope that people who do understand it are able to gain insight that they then think we gave them. When you can’t add specific value, make it look pretty and hope for the best.
The environment was quiet, purposeful, focused. None of the extraneous exuberance of Startup Weekend, with a roomful of pure hackers it’s all heads-down coding with the occasional brainstorm around the whiteboard. The minute-long summaries each team gave every few hours were all self-effacing, no-bragging “we’re doing OK, we’ve found some problems in the data but we’re working through them”, rather than the expansive “we’re building something f***** awesome and we’re going to rock your world! WOO!” of the SWPerth summaries. It’s all a bit more restrained and polite, an elegant hackathon from a more civilised age.
Sunday morning got fairly frantic. Our team had had a look at the other teams, and we were pretty sure we were stuffed. The guys with the Tonka truck were sure-fire winners if they could get it to do anything at all resembling a result. There were lots of people doing 3D stuff, maps, and quite a few dashboards. Our only hope was to make it very pretty and get it to work seamlessly, and hope we could beat the competition with sheer style and usability. So we had about 6 hours to get the 9 tiles in our display working from live data that we barely understood through a mangled Ruby gem environment that worked for only some of us. Izzy wizzy let’s get busy!
We cheated a bit: logos for TR201 and Goldfields ate two of our tiles. But the other 7 were bona-fide data displays from the live data, with further pages backing them up with drill-down detail on each metric. We were proud of it by the end; given what we had when we started, it came out good.
The preview pitch went well, Sean reckoned it was valuable, the Goldfields guy was all smiles and happiness around it. I think we finally stopped coding about 5 minutes before the deadline (that was me finally getting the rainfall data to align with the ore haulage cycle data so we could correlate rainfall and productivity; the overground part of the mine stops working when it rains – useful or not? No idea, but wedge it in there and see if anyone likes it).
We drew our place as number 15 out of the 18 teams, so we knew we’d have a chance to work out how to pitch it by the time we got there. The minus side was that if the judges had seen a lot of dashboards by then, then we’d be an also-ran unless we could stand head-and-shoulders over the others in usability.
The other teams were a mixed bag of stories. Some had very little but presented well, others had awesome solutions but they didn’t manage to get the awesome across to the judges.
My personal favourite was the team that solved the Rio Big Rock problem exactly as suggested; using computer vision. They had a working solution of spotting big rocks in trucks using the crappy video Rio provided. For Rio to have suggested it they must have looked at this themselves and not solved it – otherwise why would they present such a specific problem and suggested approach to us? So these guys (Alex, Stef, not sure who else) got this very hard problem solved in a weekend using a scalable, cheap solution that utilised footage already available. But they had the misfortune to be up against a very dramatic solution of the same problem, so got upstaged.
Our presentation didn’t go well either. Myles and Tam did OK, but ours was definitely an also-ran by the end. The judges had seen about 6 dashboards by then, and while ours looked good and worked well, it was clear that they’d seen enough actual real problems being solved to not give much credit to just a dashboard. Shame, but them’s the breaks.
My take on the winners:
I didn’t quite see how detecting vibration on a Tonka truck on a medicine ball would scale to a full-size dumptruck, though I loved the innovative approach and dramatic presentation (and the best of luck commercialising it, chaps, you’ve got an awesome product there).
I didn’t get what PitIQ actually did, so that was a bit of a surprise winner for me, though the presentation was fantastic.
And The Froys worked wonders with the Goldfields data to make it draw those lovely flows, superb presentation, but the machine learning bit got totally lost, which is a shame because the team are the wizards behind Datagami and really know their stuff about machine learning.
What I think we should do better for next time: Get some miners in the room. Sean was invaluable to us, but he didn’t know the data specifically enough to bridge the gap for a bunch of clueless web dev geeks who had never been near a mine in their lives.
All in all, this was an excellent start. We need more of these, and to refine the format and participation until we get better results. I really hope the presentations to the mining companies go well and they see the potential in more of this. A huge Well Done! to the Unearthed team for putting their time and energy where their mouths are and getting this organised so well. I look forward to the next one 🙂
The Unearthed Team take their bows. Well done chaps!
If you attended, I’d love to know what you thought of the whole thing. Let us know in the comments.