A former ISRO scientist is on his second run as a data scientist at Sahaj

From 1984 to 2008, 24 years to be precise, Ravindra Babu Tallamraju worked on several space missions at the Indian Space Research Organization (ISRO). After that, he joined Infosys as a senior researcher to work on facial recognition systems. Then, as a Senior Data Scientist at Flipkart, his task was to solve problems related to address classification, catalog management and fraud modeling.

Today, Ravindra is Head of Data Science at Sahaj Software Solutions – a software engineering consulting firm harnessing the potential of data with platform engineering, data engineering and data science.

Days at ISRO

ISRO presents a number of challenges in deterministic mathematical modeling. Spacecraft Orbital Dynamics activities involve considering the various forces acting on the spacecraft and integrating these equations of motion to predict the position and velocity of the spacecraft in future instances.

Position and velocity vectors help derive a wealth of information such as spacecraft visibility on ground stations, ground tracks, Earth shadow entry times, etc. In addition, camera modeling is used to calculate imagery scan coordinates for remote sensing missions.

“While most models involve all aspects of mathematics, such as forming and solving differential equations, geometric and numerical modeling, vector algebra, etc., the challenges of probabilistic modeling and learning automatic remain limited,” said Ravindra.

“It motivated me to look for new challenges after more than 24 years at ISRO,” he added.

Current projects

Sahaj works with his clients on purpose-built solutions involving data and ML. The customer base for AI problems is spread across multiple domains and geographies.

This offers Ravindra the opportunity to work on a larger scale. The team is involved in the development/optimization of multilingual NLP models, text summaries, knowledge graphs, natural language queries and contextual conversational engines that allow to develop multiple use cases by exploiting the engine.

In computer vision, Ravindra said, “We are working on solutions for generating video summaries, sports video analytics, information extraction from videos and images, OCR of noisy scanned documents and knowledge graphs.” Use cases for these AI models include

  • classification of damage in cars for insurance
  • identify logos, objects and
  • actions from video or images

In exploring large datasets, both structured and unstructured, the job is to extract hidden patterns in the dataset. Similarly, when it comes to statistical forecasting, the team worked on models that predict using conventional time series, machine learning, and deep learning approaches.

All of the projects mentioned above are collaborative in nature, and Ravindra leads a team of data scientists at Sahaj. According to him, data scientists are intrinsically researchers. “They like to work on open-ended but time-limited research problems,” he said.

Compete against data scientists and challenges

On the one hand, every data scientist has their chosen field, while on the other, customer projects require time-bound solutions. The need is to find a sweet spot between the two. Ravindra believes that an ideal data scientist should work on various problems across multiple fields and application areas of data science.

In his own role, he has the responsibility to focus on how to train every data scientist as a complete expert. Additionally, to instill research curiosity, the organization has a regular data science research forum where the team offers talks, conference presentations, research publications, and more. Additionally, Ravindra said, “We collaborate with some of the leading academic institutions and undertake joint research projects with expert academics,” he added.

Even though the need for data science is growing across the industry, their skills issues remain loosely defined. In addition, the team must look for several factors,

  • A data scientist must go the extra mile to understand the ecosystem and the domain before they can formally define the problem and solve it.
  • It is not enough to have math skills; therefore, the team also seeks advisory and problem-solving skills.
  • Even a new data scientist should enter the workplace with a solid theoretical understanding of NLP, computer vision, optimization, statistics, etc. and have the ability to solve problems independently and to read and understand research publications.
  • A good theoretical and practical understanding of state-of-the-art methods, including deep learning, is a good thing.

At Sahaj, according to Ravindra, any new data scientist is usually paired with experienced data scientists who can hone their consulting and problem-solving skills.

Patents are important

Ravindra rightly understands the importance of patents as he owns a few computer vision patents himself. “Associated with owning patents, we need an IP team that follows their possible infringements. Sahaj, as an organization, believes in sharing and open sourcing. As a data science team, we share our knowledge through publications, conferences, open source contributions and blogs,” said Ravindra.

However, this does not apply to any specific work deliverable or that has been developed as a solution or part of a solution for any of Sahaj’s clients.

Additionally, Ravindra sees huge opportunities for AI innovations in India. “Over the years, I have noticed a marked difference in how open academic institutions are to supporting private industry through joint collaborations,” he said.

In the industry, there are pockets that are embracing AI to a large extent, but there are startups that are still at a very nascent stage. “This requires data science leaders to create awareness of how data science could help deliver business advantage and understanding the field through interactions with business, product and engineering leaders to to identify the problematic landscape,” Ravindra concluded.

Sean N. Ayres