I am Rajeev Dutt, founder, and CEO of AI Dynamics, a startup from the Pacific Northwest focusing on machine learning, particularly in the biotechnology and manufacturing sector.
Our core product is the NeoPulse Platform, something that we refer to as an operating system for machine learning that enables enterprises to build AI solutions from inception to deployment and management. We enable enterprises to create and deploy AI solutions in weeks rather than months. One other thing - we have a great auto ML solution that truly democratizes AI.
It is exciting to see the growth of our business and the evolution of our product. Our product has already demonstrated impressive achievements. We have been able to achieve 97% accuracy in diagnosing COVID from lung X-Rays - beating the previous records, we can achieve over 88% accuracy in diagnosing tuberculosis from blood assays, and in a genetics AI project for a large biotechnology/pharmaceutical company, we were able to surpass even Novartis.
What's your backstory and how did you come up with the idea?
I am a theoretical physicist by training with a background in quantum information theory. Before founding this company, I had been working in the high tech sector for over 15 years, including companies like BBC Worldwide, Hewlett Packard, Microsoft, and Intel, and my own startup in London in the heady dotcom era. In my career, I had done a range of jobs from research in quantum optics, to working as a solution architect for enterprise computing, to management, giving me exposure to a broad swathe of the technology landscape.
I had been interested in artificial intelligence and machine learning, having read Isaac Asimov’s robot series. At the age of 13, I wrote my first AI program for a game I wrote (based on the TV series ‘Airwolf’, which probably dates me). I had read about Bayesian Networks and was fascinated by the idea that a network of probabilities could be taught to do things - in this case, ‘understand’ what the player was likely to do. It was primitive, but it did work. I studied quantum information theory because of my belief that intelligence was quantum mechanical in origin, inspired by Roger Penrose.
As I worked in the tech sector, I worked on some ML solutions and got a patent in HP, ironically using Bayesian Networks to predict network events. The thing I realized was that there was no recipe for AI - it was something that remained out of reach for most people, which severely hindered its growth. Even today with a plethora of tools, it is still something that is somewhat arcane and can only be used in its most elementary forms. I also realized that as AI improved, it would become a powerful force in the economy and development of nations and that the rich world had an unfair advantage. I was born in Zambia and spent my first ten years there, and I so I understand the potential of people but see the immense roadblocks to their success, and lack of access to machine learning will become one of the greater hindrances to ‘leveling the playing field’. Globalization has led to billions of people emerging from poverty, by providing access to technology, markets, and ideas, but AI is different - it is a subject that is, by nature, something that still requires deeper knowledge than the typical software engineer possesses. Most ML sources are open source, but the knowledge and resources needed to develop ML systems are still very much a rich world domain. I wanted to create something that would ultimately make AI universal.
Our first version of NeoPulse Studio was able to automatically solve some machine learning problems using a basic autoML solution, using just 14 lines of code. It has evolved quite a bit over the last several years to become a no-code, drag-and-drop solution. We have been used by Stanford University, Cisco, University of Essen, Dell to name a few. Next year, we will release a free version of NeoPulse that we are committing to keeping free forever. We want AI to be universal.
Take us through the process of designing, prototyping, and manufacturing your first product.
We are a traditional software house. It took us about three years to develop the core solution, which we made as a pure server application, exposing an API and a command-line interface CLI to access the API. We decided to create a custom language called the NeoPulse Modeling Language (NML), which is a simple, but Turing Complete language to significantly reduce the amount of coding for ML. It’s kind of like the SQL of ML.
The original PoC version was really to demonstrate capability and we were praised by customers, who liked the approach, and we were able to form partnerships.
Stay persistent and adapt.
It suffered some drawbacks; namely, the fact that there was no GUI and CLI effect. We changed that in 2018 to add a graphical layer. In 2020, we started to radically revise our GUI for simplicity of use. Using the most recent version of NeoPulse, it is possible to go from training data to create the ML architecture to training in a couple of minutes. We have an IDE inside NeoPulse to accelerate the process and a pure GUI so that you don’t have to write a line of code.
Describe the process of launching the business.
To some extent, we were early to market. The AI market was barely in its infancy and few AI companies were generating any revenues. It’s still early days for AI, but we are seeing greater adoption.
We were funded by angel investors from the Seattle area, which allowed us to get through to initial PoC development. We continued to get funding from angels to get to early commercialization. In 2020, we started to truly commercialize, getting past the early PoC versions of our software. Unfortunately, COVID hit and many companies started to retrench. Toward the latter end of 2020 and the start of 2021, we started to see revitalization, and now we are closing several contracts with large companies.
The most important lesson we learned was to hire slowly. In the early days, the shape of the product was yet to be defined, and in retrospect, we would have hired more slowly and perhaps aligned ourselves with a particular vertical.
We were building a generalized platform, but that’s often a hard concept for investors to get their heads around. So simplicity was important. Our product was intrinsically hard to explain to non-experts, who were confused by the reduced code size and who didn’t understand the impact. We have a simpler story to tell now.
Since launch, what has worked to attract and retain customers?
Cold calling doesn’t work. It’s time-consuming and leads to several dead ends. The most successful endeavors were through our network. We were also selected by Plug and Play to join their IoT and industrial cohort in 2018, which was a huge help. Since then, we have started to penetrate the Japanese market and even have a Japanese subsidiary - again through our network. We also worked with iPark, an incubation center in the biomedical space started by Takeda Pharmaceuticals. We have built a close relationship with them and we greatly value their partnership.
SEO has led to greater exposure and we have attracted some interest through that (in fact one of our bigger contracts). However, contact through our website is hit-or-miss.
Enterprise sales are hard and slow. It takes months to build a relationship and if a principal contact should leave the organization or be shuffled around during a reorg, all the momentum is lost and you’re back at square one. It is important to build a network of alliances in the enterprise from the most senior levels to the engineering level. The best way to start is through an introduction.
Conferences are ok and can be a source of connections, but it is important to pick the right ones. Large conferences like CES are a waste of time and money. Of course, during COVID times, conferences are less effective. Sadly, business these days is mostly over video conferences, which is certainly not the easiest way to win new customers. A pitch is less effective when you can’t see the body language of your counterpart.
We initially tried to use AWS and Azure marketplaces as a launching platform, but the pricing model didn’t work. We offered a free trial but they didn’t. This meant that customers wanting to try out our product had to pay for their use, even if we weren’t making any money on it. That is a fundamental challenge for them. At best, we see their marketplaces as a way for customers to start to scale out their solutions if they are opting for a cloud solution. In our case, most of our business is ‘on-premise’.
We are forging a strong relationship with Dell Technologies, signing an MoU with them at the end of 2020, and they have recently released a joint solutions brief that can be found here/
It is a powerful solution that can enable enterprises to grow and scale their AI infrastructure at breakneck speed and deal with petabyte datasets. We have targeted the manufacturing sector, in particular, ‘smart factories’, but our solution can be applied to the healthcare sector.
The link to the Soracom website.
We have found that there is no ultimate secret to retaining customers. Excellent customer service is the only way. Build a relationship with your customer, empathize, understand their problem and be passionate about solving their problem. Your mission is your customer’s success.
How are you doing today and what does the future look like?
We are revenue-generating but not yet profitable. In the last year, we have seen over 500% growth, and last year 30% growth over the previous year in revenues (despite COVID). We are on track to see a similar growth pattern next year, based on the deals we are closing and the current pipeline.
We plan to start marketing NeoPulse next year (we have not had a marketing campaign yet) and also plan to release a free version of our product for non-commercial use targeted at academic institutions.
All our sales are direct B2B sales today and are likely to be the case for some time in the future. Our primary market is biotech and manufacturing and we will leverage our existing partnerships and customers to continue to expand in these areas.
Long term, we will be offering a more integrated biotech service leveraging our powerful NeoPulse platform. We know that there is a demand for this as AI starts to take hold in the pharma and biotech industry.
Through starting the business, have you learned anything particularly helpful or advantageous?
Building a general platform is difficult. While NeoPulse can be applied to many industries, the challenge is that AI is not yet at the stage where customers truly understand how to use it and where to use it. Adoption has been limited.
As a result, we have discovered that selling our platform involves not only selling the core product but also the scenarios in which it can be used. Not every industry is ready for AI. Key factors that impede progress include the immature state of machine learning and the unsure returns on investment, inability to produce enough training data, and lack of understanding of the benefits of AI.
Were we to focus on a specific function or task using AI, selling the solution would be simpler. For example, if our application were to automatically analyze legal documents, then the market would be clear but considerably smaller than the potential returns on a generalized platform.
To sell a generalized platform, we learned that we first needed to identify an industry amenable to the use of AI and then develop domain expertise in that specific industry to target specific workflows and processes. We made many wrong turns before finally focusing on two sectors - biotechnology and manufacturing.
We also had to change how we presented our product to investors. The message was that we have a platform capable of solving AI solutions in almost any industry, but we are proving it in two narrow sectors. In other words, demonstrate viability in a narrow vertical but have the potential to expand to any vertical. The lesson we learned, in the end, is that, while it is great to have lofty ambitions, it is important to stay focused on a much smaller problem to demonstrate success.
What platform/tools do you use for your business?
We have used many back-office tools to handle payroll, security, source control, email and messaging, document storage, sales tracking, accounting, etc.
We run a global organization with offices in Tokyo, Vancouver, Seattle, and the Bay Area. A company our size would not have been able to do so even ten years ago, but now with HR and payroll systems spanning countries, global production systems, and cloud systems, we can manage our workforce globally.
What have been the most influential books, podcasts, or other resources?
Books that have influenced my thinking:
- Causality: Models, Reasoning and Inference by Judea Pearl
- The Innovator’s Dilemma by Clayton Christensen
- On Intelligence by Jeff Hawkins
- The Goal by Elihu Goldratt
- Crossing the Chasm by Geoffrey Moore
I am an avid reader and typically read several books at the same time. At the moment, I’m reading Code Breaker by Walter Isaacson, Army of None by Paul Scharre, and The Battle for Leyte Gulf by Comer Vann Woodward.
Advice for other entrepreneurs who want to get started or are just starting out?
My advice is simple: Stay persistent and adapt. You will make mistakes, the mistake is to throw good money after bad. Learn to cut your losses and to move on. There will also be many naysayers and so persistence is a necessary characteristic. The key is to survive long enough to achieve success.
The second piece of advice is something an old manager gave me. Don’t build battleships. Battleships are expensive, time-consuming, and become obsolete as soon as they are deployed. Start small, innovate, and adapt.
Where can we go to learn more?
If you have any questions or comments, drop a comment below!
Hey! 👋 I'm Pat Walls, the founder of Starter Story.
Get our 5-minute email newsletter packed with business ideas and money-making opportunities, backed by real-life case studies.