Daniel Situnayake is a creative engineer who believes embedded machine learning (ML) is a “once-in-a-generation technology.” Here he talks about ML’s potential and introduces TinyML, as well as a few ideal applications. We also touch on his experience as a developer at Google.  

Transitioning from Academia 

Abate: Let's start with your background. When did you first become interested in embedded machine learning? Was it at Birmingham City University (BCU), where you earned a BSc in computer networks and security? Or was it after you left university?

Situnayake: I didn’t set out to work on embedded machine learning. In fact, I’ve taken a long and winding road on my way here, with many seemingly disparate experiences nudging me towards this field! 
Daniel Situnayake on machine learning
Daniel Situnayake

 
After graduating from BCU, I was asked to stay on as a faculty member, teaching classes and consulting on behalf of the university. I focused on the topic of Automatic Identification and Data Capture technologies, which encompasses all the ways a computer can recognize and process physical objects — everything from biometrics, like facial recognition, through to RFID, barcoding, and smart cards. This was my first introduction to two things: computer vision algorithms, which at the time were not based on machine learning, and embedded systems.

A few jobs later in my career, after an acquisition, I found myself working on the nascent data science team at a consumer banking company. This was where I first encountered data science tools and workflows, and where I learned about working with large datasets. It wasn’t a topic I’d ever been exposed to before, but I found it fascinating.
 
When we started Tiny Farms, I saw my first opportunity to join these fields together, using data science tools to interpret and react to embedded sensor data in an agricultural technology setting. But it was at Google, meeting Pete Warden and Andy Selle from the TensorFlow Lite team, when I really grasped the potential of deep learning on tiny devices and it became my number one focus.
 
Abate: When did you move to California? Was it for a specific work-related opportunity?
 
Situnayake: I moved to California in 2009, the year after the global economy imploded. I’d spent the past few years taking a battery of mental and physical tests with the goal of being selected as a British Army officer cadet. I wanted to fly Apache helicopters! I was due to start officer training in May 2009, so I went on a farewell trip across the US with my best friend.
 
When we stopped in California, I met a girl. A couple of months later, and a few weeks into the Commissioning Course at Sandhurst, I decided that the life of a gentleman pilot was not for me — and that I’d be better suited to a laid-back life in California. I booked a flight, moved in with the girl I’d just met, and we got married six months later!

The marriage didn’t work out, but I’ve been here ever since. Making my transition from academia, I joined an AI startup in Los Angeles and then moved up to the Bay Area just in time for the 2010s tech boom. It’s been a crazy ride.

Insect Farming Tech

Abate: You co-founded Tiny Farms, which you describe on LinkedIn as "America's first insect farming technology company." How did that company come about? And what is the status of the company?
 
Situnayake: Like all my best stories, this one also starts with a trip across the US. Two close friends of mine had been thinking about starting a project to address food security. During a cross-country trip, they stayed in a cabin by a lake that was surrounded by big, noisy grasshoppers. On a whim, they wondered if they might be edible. After a bit of research, they learned that edible insects were actually extremely important to global food security — but that techniques for raising them in captivity had barely been explored.
 
After some initial experimentation, they asked if I’d like to join them in starting a company. I saw a lot of potential for applying the techniques I’d learned around Automatic Identification and Data Capture to automating the rearing of insects, and in using data science to guide the iterative development of novel systems.
 
I jumped in with both feet! At first, it was a part-time project, but we eventually raised money and built an industrial scale insect farm in San Leandro, California. It was intense, exciting, and more than a little stressful trying to keep millions of crickets alive while dealing with customers, investors, and research. Around five years in, I decided it was time to throw in the towel. My co-founders carried on for a couple of years, but in the end, it was too difficult to reach product-market fit, and they called it a day.
 
Tiny Farms lives on as an open-source repository of the designs and data we accumulated during seven years of operation.