SiMa.ai, the machine learning company, claims to deliver the industry’s first software-centric, purpose-built MLSoC platform. With push-button performance, the company claims to enable effortless ML deployment and scaling at the embedded edge. This is done by allowing customers to address any computer vision problem while achieving 10x better performance at the lowest power. 

To understand more about the company, Analytics India Magazine reached out to Krishna Rangasayee, founder and CEO, SiMa.ai and Harald Kroeger, president (automotive business) at SiMa.ai. 

AIM: How did the idea of Sima.AI come about? 

Krishna: We started the company with an observation that everybody was talking about machine learning at the embedded edge. People spoke of robotics, and automotive elements, etc, but if you take a long, hard look, in the past 10 years, there has not been a whole lot of ML shipped in scale. There is a dissonance between aspiration and reality. So, if you want to start a company and create a purpose-built full stack ML solution, hardware and software, we would like to address computer vision as the first problem. 

Now, four years later, we have a solution that we’re shipping to customers, delighting them with the possibilities. I think we’re changing the mindset around how to really scale machine learning and scale for the embedded market.

AIM: Harald, what was your journey like and how did you decide to join Sima.Ai?

Harald: My journey is a real car guy journey. I started off at Mercedes research, where I was later heading development of all electronics, including e-drive and powertrain. I left Daimler after 20 years and joined Bosch, which is a real tech company, more ‘tech’ than Daimler in terms of engineers and innovations. At Bosch, I was a member of the executive board and guided 100,000 people with about 20 billion of business.

Last year, in December, I decided I was going to leave Bosch to do something new. I have known Krishna for quite some time, and he asked me if I would be willing to join the board for Sima.ai. I agreed, and started the automotive business. I think the product is perfect for automotive.

AIM: How is MLSoC™ Platform different from others?

Krishna: A key technical merit that customers care about in computer vision is frames per second per watt. It’s like how much you can compute for a given power at a given time. So that’s a key technical merit people look to, and in that aspect, we are in the minimum 10x ahead of Nvidia or Qualcomm or any other competitors. 

People are surprised at how 140 people could pull off something that a 16,000-strong company cannot. How are we able to scale 10x? Number one, we do not have any legacy architecture. If you’re a Qualcomm or an Nvidia, you have a lot of legacies that you have to support, and with legacy, there’s baggage. But how far can you push technology? 

One thing that works for us is that we are completely purpose-built from scratch and it’s a huge advantage for us that we are good at solving this problem.

The second is we have taken a very software-centric approach to solving this problem. So from our perspective, machine learning is primarily a software problem. 

We’re solving machine learning inference and we know everything about the problem ahead of time. We know the network, the weights, the biases, the input data, the memory architecture, and the data flow. So what we do is, compile the problem, and we know exactly what needs to be done at every clock cycle. So, we are more on the orchestration of the data and memory management. And we do that more cleverly than most companies. 

AIM: Do you see India as an emerging market?

Krishna: We do believe India is a very big market and we are in the early stage in terms of engaging the customers. Still, we are engaged with about seven to eight customers in India. There are a lot of companies looking for a standardised edge computing platform that does a lot of different applications. It could be around factory floor automation, or security and monitoring. And so I would say, we are at an early stage, but we also have a lot of interest from customers in drones and UAVs in India as well. Computer Vision is a massive market. And I believe that India’s got a lot of innovation going – we’ll be participating in that as well.

AIM: When will India have driverless cars?

Krishna: People have had proprietary inside proprietary knowledge, and that’s been a defensible barrier for how companies preserve and retain their competency. Machine learning is breaking things apart and allowing new people to come in. There is now a jump ball where legacy is no longer a strength. So it is now open for  everybody to figure out what needs to be done, and I think India is going to take a very different approach than the classic Western one.

I visited a few companies when I was in India. I met the folks at Ola cabs, and saw the entire ecosystem of them trying to launch their own automotive future or e-bikes. There’s a lot of experimentation, and they have the opportunity of taking something brand new and really run with it. The needs of India are different from the needs of the Western world. 

Harald: A lot of times, we only think about the car being fully automotive – there is no wheel in the car, and it takes you from point A to point B. No doubt, that’s a great vision, a great marketing tactic for companies, however, to see a fully-automatic car juggling Bangalore’s infamous traffic will take some more time. 

The level of communication while driving in Bangalore is a different level of challenge altogether and AI being able to deliver it is still a distant dream. Nonetheless, if you look at the people getting killed in car accidents due to stupid mistakes, and human errors, that can be easily avoided by the introduction of AI in automative industry. AI can help avoid accidents related to human fatigue, and lack of attention etc. The cars may not get fully automatic in countries like India anytime soon, but for sure, it’ll incorporate some of the good stuff already present on the table.