Varun Khare, CEO of NimbleEdge, sat down with Neeraj Hablani, Partner at Neotribe and a member of NimbleEdge’s Board of Directors. Below is an excerpt from their discussion.
NimbleEdge is hiring across sales and engineering. If you are interested in helping unleash machine learning at the edge, reach out to us at nimbleedge.ai/careers.
Neeraj Hablani: Varun, it’s a pleasure to chat with you. Let’s start off our conversation before NimbleEdge; tell us a little bit about your upbringing and background.
Varun Khare: Thanks for having me, Neeraj. I’ve always been someone who experiments and tinkers – someone, who combines concepts and studies the outcome. This was true for me in art class when I combined oil pastels with water based on my understanding that oil floats on water – and sure enough, what a mesmerizing pattern emerged. This was also true in math when after a debate with my geometry teacher on whether two graphs would intersect or not, I created a Java program that could plot any implicit function. I’ve also had a somewhat unique curiosity. Whereas many of my peers sat for job placements, I wanted to explore uncharted waters and work on projects that were less mainstream.
NH: It seems you’ve explored uncharted waters both literally and figuratively. You’ve had the opportunity to work on initiatives in Singapore, Frankfurt, and a few stops in between. What drove you to these places as both a scholar and employee?
VK: Throughout my journey there was one primary question I was exploring – how does the human brain think so logically? Our memory, concepts, and reasoning are distributed across trillions of neurons. And yet, we merge these concepts as atomic units each acting like a brick to build a much more complex system. In Frankfurt, I went to study neuroscience and appreciate how neurons fire and activate each other for decision making. Berkeley was a continuation of these efforts where I started to work at the intersection of reasoning and machine learning. Armed by the knowledge from neuroscience, I began reverse engineering what I saw in Neuroscience to AI. In London, I was leading on-device Machine Learning at OpenMined, a non-profit focused on building technology for the privacy-aware digital age. This was one of the first open source ecosystems to successfully train ML models on users' smartphones.
NH: And was it the OpenMined experience that influenced you to start your own company or have you had the urge to be an entrepreneur much before that?
VK: My experiences have taught me that I enjoy the crossroads of entrepreneurship and research. NimbleEdge, fortunately, gives me the opportunity to explore both – the thrill of a startup while working alongside the research community. It was important for me to create a company that has the potential to scale and benefit billions of users.
NH: Absolutely. User benefit at scale is something that excites me too. One of the foundational questions we ask ourselves at Neotribe is “why now?”. Talk us through the macro environment and why the time is right for decentralized and edge compute.
VK: Cloud compute is great, it’s a trillion dollar market and a huge enabler for technology. However, there’s way too much centralization happening with the cloud. As a thought experiment, think of how much society has progressed in one decade of the main frame era versus one decade with personal computers. Whenever you decentralize, you empower people. A billion human brains improving the system together can't be matched.
My thought process to start a company was further bolstered after countless customer interviews with bootstrapped startups and large enterprises alike. They all had their fair share of problems with a traditional cloud deployment – ranging from high cost deployments to excessively large generation and consumption of data causing failing pipelines to issues related to user privacy. It was apparent companies need an alternative. So we setout to build one!
NH: To synthesize for our audience, NimbleEdge aims to move away from traditional cloud-based architecture where compute and data occur in centralized servers towards edge computing where processing happens at the edge, like on a phone or laptop. Why is this a category-creating initiative and why is the concept of distributing workloads so powerful?
VK: This is a complete paradigm shift. As smartphones became ever-pervasive, the world shifted towards doing everything digitally - shopping, learning, even dating! Smartphones coupled with the cloud have helped proliferate powerful applications. But, the world is ever changing and today, we face new challenges. First, when CloudFlare had an outage, 50% of their total requests were impacted, making cloud a single source of failure for the entire internet. Second, cloud bills are costly, somewhat unpredictable, and represent a large percentage of cost of revenue for SaaS companies. And third, data privacy is increasingly important – nobody, including cloud providers, should have access to your data.
NH: In addition to reliability, cost, and privacy, your architecture unlocks hyper personalized ML models for each user without data ever leaving your device or being sent to the server. What is it about today’s state of technology that enables this architecture today?
VK: Mobile devices are powerhouses in your pocket and thus, perfect for edge compute. My smartphone has the same RAM as my laptop and is as powerful as a desktop from five years ago. I can run a state of the art recommendation model on a $200 smartphone using less than 300MB of RAM. Further, we spend over 5 hrs a day consuming content online. We can personalize these AI models locally and fine-tune to a user’s preference. Machine Learning is not bottlenecked by hardware infrastructure or heavy GPU requirements. We can run 100 million AI models for 100 million different users. Personalization should be truly personal – and that’s the benefit we aim to unlock.
NH: What I love about NimbleEdge is that it’s so counter-culture yet so intuitive: counter-culture because you are building an approach that isn’t dependent on large cloud providers as are most modern deployments; Intuitive because you are bringing computation and data closer thus improving latency and bandwidth by definition. How do developers and business managers react when you talk about NimbleEdge?
VK: Very well said. We’ve all heard the maxim that great technology is indistinguishable from magic. Technology should be intuitive. Occam’s Razor! This is what I love about bringing compute back to a user’s device. The computation is coming back to where it belongs. During conversations with customers, developers, and business managers, we’ve repeatedly heard that Edge Computing is something they have asked for but the infrastructure previously was unavailable for them. That’s what NimbleEdge solves.
NH: You brought up a few stakeholders – customers, developers, and business managers – let’s talk through the value proposition for both your customer and your user. Walk us through 3 specific examples how NimbleEdge impacts customers and users globally.
VK: Sure. Today, every content over-the-top platform or video recommendation engine is fighting for the viewer’s limited attention. With us, companies can create individualized models for millions of users with models trained on individualized user data. This unlocks hyper-personalisation and a more relevant user experience. Every social media company struggles with humongous cloud costs proportionate to their Machine Learning Operations; this can be done at a fraction of the cost with edge computing all-the-while offering enhanced privacy for users. Lastly, FinTech companies are excited about the latency benefits NimbleEdge provides. They often struggle with deploying Fraud Detection Algorithms which tend to increase the latency which can drive their customers away from their platform. Every millisecond matters in ensuring only authentic transactions make it through.
NH: To summarize, you empower customers to build digital services that aren’t reliant on centralized cloud servers and users to enjoy products that are more performant, lower latency, and also have higher privacy standards. At this point, you’re on millions of devices globally. What’s next?
VK: A billion devices! We are creating an entire ecosystem for edge computing. We envision this ecosystem as the foundational technology orchestrating the computing infrastructure for the decades to come. As we say, liberation for devices, privacy for users, and profitability for companies. We won't stop till we reach there.
NH: Thanks for your time, Varun. As we wrap up, do you want to share how a company might determine whether they should try NimbleEdge?
VK: Thanks for having me, Neeraj. If you are looking for hyper-personalized models which will save you $10+ million per year in infrastructure cost and generate another $10+ million per year in revenue then you should talk to us.