Dr Yogesh Karpate has led numerous R&D and industrial projects and managed teams to develop robust solutions for diverse industries, including oil and gas, defence, insurance and healthcare. 

Before Chistats Labs, he held significant roles in R&D at Children's National Medical Center USA, contributing to innovative healthcare projects and collaborating with various international academic institutions.

INDIAai interviewed Dr Yogesh to get his perspective on AI.

How does an electronics and communication engineering graduate like you get into AI?

AI is a multidisciplinary domain encompassing neuroscience, probability and statistics, linear algebra, information theory, and mathematical optimization. Contrary to the common perception that AI is solely a branch of computer science, it extends beyond this. The field of electronics and communication, particularly in signal processing and communications, covers many of these aspects. Therefore, transitioning into AI was seamless from a theoretical and practical standpoint.

Tell us about your research at the French National Research Institute for Computer Science and Control. Tell us about your research contributions.

My thesis centred on identifying and localizing multiple sclerosis lesions in multi-channel MRI brain images through robust statistics and weakly supervised methods. The project faced several challenges, including a limited dataset and the issue of image intensity distributions that varied significantly. Moreover, the lesions lacked consistent shape, volume, or intensity patterns.

The first significant contribution of my work involved standardizing these intensity distributions across images to similar ranges. To achieve this, I developed an innovative method based on gamma divergence and a maximum likelihood approach tailored explicitly for intensity normalization.

The second essential contribution was the development of a technique for lesion identification using a one-class learning method. Both these methods were subsequently implemented in clinical practice through the French Life Imaging projects and medINRIA, a tool for medical image analysis.

What were the initial challenges you faced in the Chistats Labs?

Since I had not gained exposure earlier, my primary challenge was shifting from an academic to a business mindset. As Chistats Labs began to expand, scaling the company to accommodate growing demand presented difficulties. Additionally, securing new business and leads took a lot of work, as customers often preferred to work with familiar or previously engaged companies. However, this issue was mitigated by adopting a partnership model. Ultimately, all these challenges were successfully overcome.

What inspired you to start the Chistats labs?

Having worked on AI/ML applications at research and development-driven institutions in France and the USA, I was deeply inspired by their emphasis on translating R&D into practical solutions, which I greatly admired. Moreover, my wife, Vanashree, consistently supported and motivated me to pursue entrepreneurship. Upon returning to India, I observed a persistent gap between the demand for AI/ML products/solutions and the available skills needed to deliver these tasks. Contrary to the traditional IT industry's belief that it was just another software development task, I recognized it as a distinct and complex field. This realization led me to embark on my entrepreneurial journey in 2017.

Tell us about some exciting projects you are doing at Chistats Labs. 

Our team has successfully contributed to over 25 diverse projects and solutions in the US/Europe and Indian markets, all centred around AI/ML and data engineering. Among these, several standout initiatives illustrate our capabilities. For example, we engineered and implemented an AI-powered early warning system capable of forecasting malfunctions in specialized Oil and Gas pumps three days in advance. This system has enhanced operational efficiency and resulted in approximately $3 million annual savings. In the financial sector, we've created a product for options derivatives trading that generates substantial returns. This product leverages AI to predict price movements and devise effective trading strategies accurately.

Additionally, we developed an underwater threat assessment system employing acoustic signals to enhance situational awareness for the Indian Navy. This system is pivotal in strengthening naval defence capabilities. 

We have developed several AI-based platforms for the Defence Research and Development Organisation (DRDO). These include advanced systems for intelligent image analytics, evaluating AI algorithms' reliability, and monitoring aero gas turbine engine health. 

Each of these projects reflects our commitment to delivering niche, innovative AI and data engineering solutions, showcasing our team's expertise and the impactful outcomes of our work.

What advice would you offer to someone looking to get into AI? What is the best way to advance?

It's essential to thoroughly practice basic data engineering for preparing data for downstream machine learning pipelines. Engaging in various ML challenges, such as those on Kaggle and Drivendata, and prestigious conferences like KDD, ICDM, and VLDB can be immensely beneficial. Participating in just a few challenges annually can significantly elevate your expertise and confidence, propelling you into a higher orbit of knowledge and skill. 

Additionally, it's essential to determine the specific type of data that interests you, as this choice can significantly influence your focus area. For instance, if you're drawn to text data, you would naturally gravitate towards Natural Language Processing (NLP), a field distinct in its techniques and applications. Conversely, if you're interested in time series data, particularly from IoT devices, you'll find that it encompasses numerous signal processing applications. On the other hand, image data requires a different approach, often involving methods unique to image processing and computer vision. Meanwhile, graph data is extensively used in network analysis, as seen in platforms like LinkedIn, where it's applied to understand and leverage complex relationship networks. Each domain has unique characteristics and challenges, necessitating specialized knowledge and skills.

What quality or skill do you expect from the first-year students who wish to join your company?

Having a solid foundation in programming and core academic subjects is essential. I've noticed that first-year students often rush into complex topics like neural network backpropagation. Yet, they find it challenging to explain fundamental concepts like mean, standard deviation, recursion, etc.

What are some essential research articles and books that inspired you?

For basics and concept illustration, Pattern Classification by Heart and Duda and Bayesian Reasoning and Machine Learning by David Barber

These are excellent books

• Stacked generalization by David Wolpert for advanced ML
• Deep Learning by Ian Goodfellow is an excellent text to follow for deep learning.