Karim Aly, CEO of Noze – Interview Series
Karim Aly is CEO of Noze, a Canadian AI startup that has developed the world’s leading technology to digitize the sense of smell. He is focused on executing the company’s vision to transform healthcare by empowering machines with the ability to smell.
Prior to Noze, Karim established one of the first startup studios in Canada in affiliation with one of the country’s largest universities. Earlier in his career, he was an active entrepreneur in emerging markets, having founded multiple technology companies that scaled to over 20 countries across the Middle East and Southeast Asia.
The spark for the idea for digital olfaction was initially conceived in 2014, could you share some insight from these early days?
Of course. It was really a function of the natural curiosity of our founder and CTO – Ashok Prabhu Masilamani – where he was driven to understand why we had successfully digitized sound (microphone), digitized vision (camera), and digitized touch (haptics), but not smell. As he peeled back the layers, he began to understand the key failure points that had held us back in the pursuit of digitizing smell. As a career scientist, these learnings became the cornerstone of Ashok’s vision for a new startup; one that would develop a platform that could truly bring odor perception to the digital world, and with that, Noze was born.
The company spent the next 6 years innovating and perfecting the world’s most advanced digital odor perception framework that has solved for real-world odor detection and tracking. While the technology clearly has potential applications in a variety of areas from air pollution to law enforcement, we have chosen to focus on applying our digital olfaction platform exclusively within healthcare. In fact, we just announced a $1 million grant from the Bill & Melinda Gates Foundation to develop an AI-powered healthcare breathalyzer that can detect infectious diseases like Malaria and Tuberculosis through the odor biomarkers (Volatile Organic Compounds) in the breath. This will be a game changer for millions of people.
In 2015, NASA’s Jet Propulsion Laboratory (JPL) had a technology that matched your team's vision. What was this technology and how did your team secure this patent?
In 2014, NASA’s Jet Propulsion Laboratory had developed an innovative “digital nose” technology to detect multiple vapors/gases in orbital vehicles in space. NASA was focused on testing this capability on the International Space Station (ISS), which is a far more arduous environment to “smell” vapors compared to down here on land. We saw huge potential in their early learnings, and so we decided to accelerate our journey by securing an exclusive license to the six patents held by JPL in the digital nose space. Since then, we have radically evolved and improved on JPL’s digital nose technology by adding proprietary layers of aroma data engineering and perceptive AI algorithms, to launch the world’s most powerful digital odor perception platform.
What are the different machine learning technologies that are used in order to produce a unique digital scentprint?
Producing an interpretable digital scentprint actually involves far more than just machine learning. At Noze, we realized early on that digital olfaction needs to be viewed as a framework, one that is similar to a mammalian olfactory system. In mammals, the front end of the olfactory system is a diverse array of olfactory receptors. In order to emulate these olfactory receptors, we built a sensor chip with a diverse array of chemical receptors. When an odor is introduced to the mammalian olfactory receptors, they produce a unique neural code, and in a similar fashion, when an odor passes over our chemical receptor array, it produces a unique “digital scentprint.”
The sensory front end of the digital olfaction framework is just the tip of the iceberg. It is backed by a cloud-based, well-curated digital odor library and a chemically perceptive AI engine. The magic happens when all the pieces work together in harmony.
Could you discuss the algorithms that are used to then interpret scentprints?
In order to interpret an odor, we have to create a dataset of digital scentprints for that odor. We discovered that the odor dataset constructed from the Noze sensor chip contains rich chemical semantic information represented in the form of manifolds. In the world of computer vision, using manifold learning techniques is a popular approach. However, unlike computer vision which is a data abundant domain, the world of digital olfaction is data scarce. So our AI toolbox applies a variety of novel approaches such as meta-learning, few-shot learning, and manifold learning on our purpose-built odor datasets.
A real-world digital scentprint of an odor would contain all the associated background noise that would typically interfere with correct interpretation. This is why our proprietary datasets are carefully curated, built using a combination of data points representing background odors (noise) as well as data points representing the odor itself. This allows our AI algorithms to be trained to recognize and reject the background noise, while correctly interpreting the incoming scentprint.
Could you discuss the Noze cloud-based platform and the process for adding new scents and how large is the library of scentprints?
Our cloud-based IoT platform hosts the digital odor library and perceptive AI engine. Our library is made up of two types of datasets; one that is actively engineered to create scentprints for selected odors and backgrounds, and one that is passively created from the continuous sampling taking place by devices in the field which contain our sensor chip. These passively sampled scentprints are curated and stored in our odor library so that they can be referred back to and matched with odors that the platform may learn in the future. Given that our platform is connected to all our devices in the field, we have also developed powerful network effects, where there is a continuous, collective learning process between devices. In other words, one device can learn to interpret a new odor from the learnings acquired on an entirely different device.
We have made a fundamental decision to focus on building high-quality scentprints which can enable meaningful use cases. Our belief is that success in digital olfaction is not merely a numbers game, but rather will be anchored in the economic and societal value that can be unlocked from the underlying odor library. That said, our proprietary library today contains over 100 well curated odor scentprints, powered by nearly 100 million data points.
What are some of the different use cases for digital scentprints in manufacturing?
One can easily begin to envision how almost any industry could derive massive benefit from the digitization of the sense of smell. In manufacturing, there are some clearly valuable use cases, particularly those related to improving safety and ensuring regulatory compliance. Imagine being able to detect a burning wire in your machinery just from the odors being released and as a result having the opportunity to stop operations before a fire breaks out, or imagine if you could continuously track a collection of by-product vapors to identify the moment their concentration rises above the HS&E threshold in order to vacate and vent the area.
Our unique capability to differentiate odor signals from background noise is what enables us to determine that the odor is in fact coming from a wire that is burning, and not for instance, from cigarette smoke or a hot cup of coffee. Avoiding false positives resulting from other “background” odors is critically important, and one of the biggest challenges, to successfully commercializing a digital olfaction platform.
How is this technology currently being used when it comes to food?
While our technology is not currently being used in the food industry, there are many potential applications across the food supply chain where it could be deployed. As an example, let’s take a look at food freshness. What if your refrigerator could detect which foods were placed inside and then predict how much time was left before each one spoils? This same solution could also be applied to grocery stores and restaurants, which along with homes, collectively account for over 80% of the food that goes to waste every year – a $400b problem in the United States alone.
From a completely different angle, digital olfaction can also help automate the cooking process by tracking the aroma of a dish or recipe from beginning to end in order to cue the chef (or automate an appliance) with instructions on what to do every step of the way. We actually built a demo where we trained our AI on the complete cooking process of a chicken breast on an indoor grill. We were able to cue the user on when the grill was adequately heated in order to add the chicken, when to flip it, and when to remove it from the grill, in order to end up with a perfectly cooked chicken breast.
One interesting use case is in detecting viruses, could you specify how this works?
The human body emits certain odor biomarkers, or Volatile Organic Compounds (VOCs), as a physiological response to infection. This phenomenon however is not limited to only viral infection. These VOCs, which can be emitted from either our breath or our skin, can indicate the presence of various clinical conditions or diseases. If you think about a “health breathalyzer” that can, with a single breath, potentially detect Malaria, Tuberculosis, Diabetes, and other conditions at their earliest stages, you can easily begin to appreciate the impact our technology can have on the ability to take timely action and improve patient outcomes. It is precisely this vision that we are working on right now with multiple partners including the Bill & Melinda Gates Foundation and The Montreal Heart Institute, among others. As a company, this is where we found our sense of purpose, and we could not be more excited with both the work that we are doing, and the meaningful impact it could have.
What is your vision for the future of digital olfaction recognition?
Noze’s Digital olfaction platform is a powerful tool that has digitized the sense of smell. In the last 8 years, we have perfected this technology to work outside of controlled lab environments. We have built several odor detecting or tracking solutions for everyday scenarios, where our solutions have worked robustly despite the challenges associated with each. Today our goal is to apply this technology to elevate human health to a completely new level. We have barely scratched the surface in terms of what we can interpret from the volatiles that are continuously emanating from our breath and skin. We believe that our platform can dramatically alter the healthcare status quo by digitizing these signatures and correlating their presence to various health conditions. That said, detecting odors from human breath and skin is not without its challenges. The volatiles of interest are usually present along with confounding backgrounds including the presence of exogenous VOCs, higher temperatures, and condensing humidity. Each of these characteristics can affect the detection accuracy, which makes it particularly challenging to build a reliable and scalable solution.
Accordingly, our vision for digital olfaction has always been unambiguous: to deliver a scalable solution that works robustly and reliably in the real world, not just in the lab. It is only then that we can truly enable ubiquitous access to screening and diagnostics that will help save lives and improve health. And today, we are on the cusp of delivering that to the world.
- Tech Leaders Highlighting the Risks of AI & the Urgency of Robust AI Regulation
- NVIDIA: From Chipmaker to Trillion-Dollar AI Powerhouse
- Laura Petrich, PhD Student in Robotics & Machine Learning – Interview Series
- Liquid Neural Networks: Definition, Applications, & Challenges
- Patrick M. Pilarski, Ph.D. Canada CIFAR AI Chair (Amii) – Interview Series