We aim to become US $1 bn company by 2027: Rohit Kochar, Founder, Executive Chairman & CEO, Bert Labs

  • August 03, 2023

In an exclusive interview with Pravin Prashant, Editor, Indian Chemical News, Rohit Kochar, Founder, Executive Chairman & CEO, Bert Labs shares his views on market size of factory automation, company’s performance, clients signed, funding, IPO listing, manpower addition, ensuring error free plant operations, managing brownfield capacities, and new areas of research.   

How has Bert Labs performed? 

In terms of Bert Labs performance, it is important to evaluate it from the perspective of Bert Platform Solution. For a deep tech company, it becomes not so prudent to just focus on our financial performance. We haven’t audited our results, but I can say that the first nine months of FY 2022-23 was slightly north of Rs. 100 crores on revenue with around 25% EBITDA. More importantly, the Bert Platform Solution has been deployed by 30 clients across sectors. This is creating digital transformation infrastructure and bringing business transformation. We commit numbers in terms of energy efficiency improvement, reduction in carbon footprint, and production efficiency improvement.

We help to improve clients on overall equipment effectiveness and quality improvement. The other way of measuring efficiency improvement is through intellectual property that we are creating since one of the pillars on which we are building Bert Labs is research and innovation. We have Digital Twins built up for the electrolysis unit, HCL synthesis rectifier and on top of these Digital Twins, we have Reinforcement Learning (RL) agents which bring in the optimization. Similarly, in the soda ash manufacturing, we have Digital Twins created for ammonia recovery, ammonia absorption, and carbonization. Again, we have Reinforcement Learning agents for optimization.

What is the number Bert Labs is aiming for in FY 2023-24?

So, in terms of numbers, Rs. 500 crore is what we are aiming for. On a conservative basis, we are reporting Rs. 300 crore but keeping the order book in sight, Rs. 500 crore seems to be very realistic. We will continue to grow at 200% plus year on year and plan to be a US $1 billion company by 2027 and then file for IPO.

Would you talk about your clients?

There is a Rs. 5,000 crore specialty chemicals company which is our client. We have been focusing on caustic chlorine as well as other bulk intermediate products. As per the clauses of agreement signed with them, towards the end, we will be deploying Bert Platform Solutions in their power plant and all the manufacturing plants. Then there are two soda ash manufacturing companies which are our clients. We are focused on ammonia recovery of ammonia absorption and carbonation to begin with. We will look at other integrated unit operations, including power plants. Besides this, we are focused on API manufacturing. So, one of the largest business conglomerates has speciality chemical manufacturing as well as API manufacturing.

Are you adequately funded or are you looking at raising funds in the near term? Are you looking at international or Indian listings?

We are adequately funded and that is what differentiates us from other early-stage companies. From day one, we have been relying on ourselves. Obviously, we started with bootstrapping this company but then we started to focus on revenues and profits. Every programme of ours and every deployment is profitable. That's why in our sixth year of operations, I spoke about profitability, which is around 24-25% at the operating level and this is unheard of when it comes to early-stage companies. So, a lot of internal accruals is what we use not just for working capital requirements, but also for investment in research and innovation. But whenever we have required external funding to fuel our growth as well as our research and innovation, we have brought in funding.

As I speak with you, we are closing a US $10 million fund through family offices. And a year from now, we would close our large global institution round from a company that is one of the largest energy players globally. Along with them we will take this company for a US IPO. So, we are looking at three rounds of funding, US $10 million at the end of August; a large global institutional round which will be for US $25-30 million; and then another round of US $50-70 million.

Would you talk about your setup in India? What's your plan on manpower addition?

One of the biggest assets as well as biggest challenge is human capital. We are currently 50 odd people, growing 3-4 people every week. Ours is a pure product platform play. We're not dependent on people whom we bring on board. By the end of this year, we plan to add 150 plus people. In the next four years leading to the US IPO, reaching 2,000 plus people is our focus. The strength comprises largely research and innovation professionals from computer science, software development, distributed computing, embedded systems, and hardware as well as software. Then there are people from wireless sensor network background. 85 to 90% of our team strength comprises researchers and scientists from premium institutions and many of them are PhDs. Along with them, we have product management leaders. So, ours is not a sales driven go to market strategy but a pure product management driven GTM strategy.

We plan to set up Bert Labs Research in 2024-25 in Bangalore, West Coast, and Germany. Bangalore and West Coast will focus on Artificial Intelligence (AI) and the research team will not have any pressures of client deliverables. Germany Research Centre will focus on everything around the Internet of Things (IoT) like IoT devices and IoT powered wireless sensor networks. This also will be a major part of the 2,000 plus team which we would build globally.

We are starting with presence in APAC, Middle East, US, UK, and Germany this financial year and global expansion will continue. We are in the process of hiring locals in the Middle East and other geographies. That is how we aim to build this team globally.

You have a baseline model and then you build up on it to improve the processes. How do you ensure error free plant operations?

These are highly constrained process environments and if there is any deviation in fact in case of API, it leads to FDA questioning and even cancelling of license. One of the Hyderabad based companies have engaged us just to focus on deviations and bring Rs. 50 crore revenue to them.

For the chemical and speciality chemical manufacturing, we do multiple things. We first understand the underlying physics and chemical engineering principles, we just don't rely on operational data of these unit operations or these equipment operational data from the DCS - PLC network, we also do simulations around physics and chemical engineering. That helps us to understand the unit operations of a particular equipment at the design level and at the component level. Then we integrate that with the historical data on which deep neural network models are trained. So, when we do that, the Digital Twin which gets created, exactly replicates the real plant environment. We hand over these Digital Twins for our clients to play with. The Digital Twins powered dashboard that we provide them can be operated from anywhere. Prediction accuracy is anywhere from 94-95% going up to 99.99% in a few cases.

With our Reinforcement Learning philosophy in action, we first train our Reinforcement Learning agents in an offline environment, and that training goes for millions of runs. Millions of episodes are called and that millions of episodes coincides with training equivalent to the next 10 years. And once these Reinforcement Learning agents start converging, meaning every time action is taken and the output which gets generated is the output which the plant is looking for, at the same time, the energy, the power consumed, the electricity consumed by the electrolysis unit also comes down. That is the time we decide in consensus with the client that the Reinforcement Learning agent is ready to be deployed in the real plant. That's how we make sure that all the operating conditions are met where the Reinforcement Learning agent is deployed to perform with respect to energy efficiency improvement, production efficiency, and quality improvement.

Now you have Digital Twins and a simulator simulation platform but if there is any exigency, how do you manage them? 

Our training happens on historical data and the exigency which the plant has gone through is anywhere from one to five years. Our neural network model or reinforcement minimal learning model has learned and has adequately taken care of meeting those exigencies and yet the Reinforcement Learning agent gets trained on a Digital Twin which has a power to predict at an accuracy going up to 99%.

We have a four-pronged control strategy. AI controls which get executed on  the cloud from a remote server. Then AI control gets executed from an on-premise server and the third level of control is on our edge computing device. So, we have two edge computing devices - Bert Titan and Bert Aksh. AI controls run on these IoT devices and are very close to where the equipment is placed. The fourth level of control is on our controllers, Bert Minnie and Bert MinnieComm, the two controllers were PID control runs. If for some reason, the edge computing device is not able to activate controls through our controllers then the PID control gets executed. For some reason if the premise server is not able to execute controls, then the edge computing device actuation control comes into picture. For some reason, if the cloud-based AI controls don’t get executed then either there is an on-premise server or the edge computing device. So, there's a business continuity which is maintained through these four prong control strategies.

It is easy to manage greenfield operations but how do you manage brownfield facilities?

In a greenfield project, we integrate Bert MinnieComm with thousands IO points, digital analogue, hard and soft IO points. In a chemical plant, we are working with, the IO points are anywhere from 2,500 to 5,000. Now in the brownfield plant, we integrate Bert MinnieComm that either integrates with the Emerson PLC or a DCS or Allen Bradley, Rockwell, Siemens and this way with other PLC and DCS servers and controllers.

This integration happens through standard industry communication protocols like OPC UA, OPC DA, BACnet etc. Bert MinnieComm in a brownfield project captures all the data points which we need and from Bert MinnieComm, the data gets transmitted through wireless sensor network to Bert Titan which is a base station of our wireless sensor network. And Bert Titan aggregates data and AI computation gets done on it. Bert Titan also pushes data to the on-premise server and to the cloud server where we have Bert Nova. Our solution works seamlessly, plug and play, both in a brownfield environment as well as in a greenfield environment. Some of the control levers which are manual, some of the control levers which are digitized, both become part of Bert Platform Solution and then compute happens in real time.

What is the market size of factory automation globally and in India?

The size of the market could be US $3-4 trillion, corroborated by McKinsey studies but we don't rely on the size of the market because digital transformation is and should be ahead of physical infrastructure where a chemical or API manufacturing or any other manufacturing plant in any sector should focus today. My advice to all my CEO friends is that if you are laying out a greenfield project, please have the digital transformation strategy executed before you are laying down the physical infrastructure of your plant and all along with it as it is fundamental and therefore critical. The other way of looking at it is sustainability and decarbonization. Today every Chairman, CEO, Managing Director in developed and developing markets, in chemical, pharma, and other sectors like cement have their mandate for zero carbon and carbon neutrality.

Whom do you consider as your competitors?

We come across Allen Bradley, Rockwell, and Siemens. We also come across companies like AspenTech, Star CCM and 1D, 3D, and CFD simulation companies; however, they are more of our collaborators. When I am talking about first principle chemical engineering simulations, our research scientists use AspenTech as a tool and they do all the chemical engineering-based simulations. When they do neural network simulations in parallel, our physics face modelling is done using Star CCM and other tools like GT suite etc. Our competitors are largely hardware companies but they do have software platforms also which are largely around monitoring and which are also largely around rule based control logic or PID control logic. We end up sitting on top of the control logic of Emerson, Siemens, and Yokogawa at our client sites and obviously one control logic has to be actuated. So, it’s Bert Optimus and Bert Geminus Control Logic which gets actuated. It overwrites the existing control logic from these technology companies.

New areas of research that you are working with respect to smart factories?

There are several of them. I am bucketing these into two areas. One is the area of sensing and the second is the ability of the algorithms to find correlations between those thousands of parameters and variables. Bert Platform Solution is at the intersection of these two. So, broadly speaking, our research will focus on how we can improve the capability of Bert Maximus IoT devices and their ability to sense capture data points. We have one image processing device called Bert Aksh which integrates digital camera and thermal camera. Now the processing board is all Bert patented but the thermal camera and the digital camera gets integrated as a third party. So, the future of sensing is non-intrusive and Bert Aksh is enabling it to sense some of the parameters. On the AI side, it is the power of algorithms and neural networks. The Reinforcement Learning that we have created is a combination of two very powerful neural networks. So it is, real time and every second, every split second, it is able to find correlations between thousands of variants. This is where we would focus heavily on research. Just imagine how powerful that algorithm will be.