Naysan Saran is Co-Founder and CEO at CANN Forecast, which uses artificial intelligence to help governments and businesses make better water management decisions, reduce operating costs, and better understand their impact on the environment. Previously, she was Scientific Programmer at Environment Canada.
We share this story as a demonstration of AI being used for good. We believe it’s important to show people what it looks like to build ethical AI that improves the world for everyone without downgrading, polarizing, or discriminating against humans.
The book that changed her life – A Tour Of the Calculus
I just stumbled upon the book at the library. I was into books and this one seemed interesting because I didn’t really know what it was going to be. Is it going to be like a novel? Or just a plain blank history of math? I was curious. It also was a small book, so if it wasn’t interesting at least I could finish it quickly. But it changed my life completely.
The author really makes math interesting by asking: “How can two people like Newton and Leibniz come to discover calculus at the same time without even knowing about each other’s work?” So it raises the question of whether math is invented or discovered. Because if it’s invented , that’s weird. If it’s discovered then it might have something in common with all the other astronomical discoveries that have been made by multiple civilizations at the same time.
I also liked the author, he was really funny. He was talking about Madonna, and the personalities of Newton being very cold and not really fun to talk to. On the other hand, Leibniz was always eating stuff and a little fat. So it makes them likable, right? It’s no longer about the name of the theory – it’s actually about the people who actually lived and enjoyed life sometimes.
The hackathon that pushed her into using ML to predict water quality
I was working at Environment Canada as a scientific programmer, using statistical modeling to predict storms and snowstorms. At the same time I was taking two math courses per semester just for fun since I had a passion for math by then. Then one of my colleagues came and said, “Do you still have spare time during the summer?” And I did, because I didn’t have school during the summers so I only had work – so I have all my weekends and my evenings free basically. So I said yes. He said, “Oh there’s this hackathon hosted by The Gaspé Beaubien Foundation.”
Their goal is to raise awareness around water. So each year starting in 2015, and now it’s ongoing, they (AquaHacking) have this six month hackathon. It’s not like a 48 hour hackathon where you just give it your all for a short period of time; it’s 6 months. They train you, and give you all sorts of training, including how agile methodology works, how to go and ask for clients, how to pitch, and how to think of the business value of your solution.
So it’s more than just the hacking part, because their goal is to have something, a tool that you can actually use and sell by the end of the hackathon.
How they disrupted decades-old methods by the end of the hackathon
My co-founder is a researcher in water quality and he was finishing his PhD. I was a scientific programmer researching water quality. He described the problem to me: right now, in order for municipalities to be able to tell if a body of water is safe for swimming, they have to sample it, send the sample to the lab, and then 24 hours later it comes back with the results. And if the water was contaminated, that doesn’t really help because that was yesterday – people already swam in that water. We want to be able to predict and prevent problems from arising in the first place.
People have been doing things a certain way for decades. We’re saying there’s a better way now. Once you think about it you realize water quality is a function of environmental factors like precipitation, temperature, humidity, sewage discharge into the river. The city of Montreal gave us that data. Then we created a machine learning model that was learning from the past. By the end of the Hackathon our solution was already more effective than what the city had been using. So that’s how we won the hackathon, and the city of Montreal saw that we had something that they could use. They gave us our first contracts.
The unexpected success of their project, and why she loves startup life
I was just happy to be doing something meaningful with smart people really. I never thought about, even when they say, “Oh you have to pitch your solution. You’re building a startup.” I was like, “Yeah, sure. It’s never going to happen, I’m going back to my government job, nine to five. My pension is already pre-planned.” But no regrets really. I learned a lot.
Also my threshold for stress got a lot higher than before. It makes you older in your head, more mature a lot faster. That’s because you go through so many ups and downs, but you learn. I’ve learned like in 10 years in public service, really. I would jump in right away. If I knew that this was going to happen, it would have been 100% I would sign off yes. For sure.
The root cause of Montreal’s poor water quality
The most important data for Montreal is precipitation and sewer overflows because the city doesn’t really have a choice. The sewers are combined, precipitation and sewers at the same time. When it rains too much, it all goes to the water treatment plants.
The problem is that the water treatment plant has a maximum capacity, so when it rains too much the water treatment plant can no longer absorb all the water coming in. So some of it has to be diverted to the Saint Lawrence River and that creates contamination, since it’s sewage. By the way, all the medication that people take goes back into the sewer. The sewer gets dumped into the river and then it just floats there.
And this isn’t only in Montreal, many older cities in North America have the same problem. We implemented the model in the city of Lévis as well. Over there they have tides, so it’s different: the inputs we consider depend on the site.
What exactly CANN Forecast’s InteliSwim model predicts
For now, we predict the only water quality metric that’s used by the Ministry of Quebec. It’s E. coli. We know that a lot of other stuff are related to water quality changes, but because they’re not used as an indicator we’ve never had a request from customers to model them.
But the metrics also depend on the purpose of the water. Bacteria and viruses are the most important ones for swimming. Besides that, minerals and pesticides are important to consider for drinking water. And for irrigation, you need to consider all of those.
How their InteliPipes product helps prevent water main breaks
So we have two products: InteliSwim and InteliPipes. The latter is also a predictive model, but for water main breaks. So we basically tell cities where they have to fix the pipes before they start leaking. A basic day in our lives is we do a lot of data cleaning because municipalities, and pretty much everybody who’s trying to do AI, they have gathered all sorts of data, but the data is not completely clean. It’s not really formatted. Sometimes you have Excel sheets, sometime you have Access documents, sometimes you have PDFs, sometimes you have text files. So 70% of our tasks is really data cleaning and formatting to put them into a proper format that our models can use. Then you remove the outliers, etc.
We’re collaborating with a research center for urban infrastructure in Quebec to develop a model that can detect if the data set has mistakes. Currently, we only have these two cases for predictive modeling. The first is predicting levels of E. coli. using environmental variables like precipitation, temperature, and humidity.
The second is predicting pipe breaks, which is a lot more complex. You have to take into account the network structure of the water network, how old the pipes are, which material they’re in, the pipe history, and if you had problems with pressure variation within the network. It’s tough, but in the end we’re predicting catastrophes before they happen.
Their biggest struggle right now: project overload
The biggest struggle right now is that we’re working on our IntelliPipes solution for municipalities, but at the same time we have a lot of R&D going on with McGill, the Montreal Institute for Learning Algorithms. We also have the consulting contracts that we do. The team is small so everybody’s a bit tired. I think we have to become more efficient at doing what we do and automate as much as possible.
We are seven right now. But I think the workload is too much I would say for the team. So we are forcing ourselves to become very efficient before we scale. My co-founder and I are gambling that if we managed to do that much with this team, then as we scale we’re going to be able to do a lot. But if we just keep adding people and not try to increase our efficiency beforehand, it’s just going to keep being too much work as we go.
What you can learn from their innovative financing solution
I think what we did with our InteliPipes product might be interesting for other startups to consider. The government can pay up to 75% of the development for deep-tech R&D projects. But you have to come up with the 25% yourself. So we asked our clients (the municipalities) to pay for that 25%. So basically they pay for a certain part of the development. We use that, we leverage that to get whatever else we need, the total 100%.
Then we develop the product with these municipalities that paid. We add the features that they need, and once it’s out we are giving them a discount for the first five years. So it’s another way to finance your development and make sure that you’re doing it with your future customers. So yeah, that I think is interesting. You don’t have to raise capital or dilute your equity.
We didn’t want to give up too much equity. So I guess we really needed to become creative. Like all the grants, all the government grants, as you raise you always have to bring a percentage. So we said, “How much do we need? All right, what’s 25% of that amount?” Then we reached out to 25 municipalities across eastern Canada and nine of them said yes.
Looking back, their biggest waste of time
Saying yes to too much. I think as a startup you’re really tempted to accept all sorts of contracts and collaborations and projects because you’re just happy that somebody is interested in working with you. But I think maybe with more experience we would’ve been more intelligent in how to prioritize our time better. I think we said yes to too much. Now we’re learning to sort our priorities. But it’s still difficult because you don’t want to let anyone down. But at the same time, if you spread yourself too thin you’re not winning any battles.
What’s next: InteliPipes across 9 cities, and algae blooms
Hopefully we get that InteliPipes product out and into the hands of the nine cities we’re developing it with. We’re looking forward to seeing how much safer it’s going to make them in 2020. Then we have a couple of other interesting projects coming up. Something that has to do with algae blooms.
But I can’t really tell you much about it. What’s important is that we want to continue developing environmental products that prevent unfortunate events and have them be used by as many cities as we can.