“Campaigns like ‘AI for India,’ computing platforms like AIRAWAT propel India toward an AI revolution,” shares Sudeep George, VP

0
29

With digital emerging as a go-to choice for the whole economy, the Technology sector in India is enjoying a sweet spot among all sectors with rapid tech advancements. Today, technologies like AI, ML, IOT, Robotic Process Automation have become face of the tech industry and are shaping the way we live. However, what has gained massive eyeballs over the last couple of years in AI/ML. To decode some insights on how AI/ML has progressed, trends in artificial intelligence and what role does data-cenrtic AI plays in building a Custom AI system, we reached out to Sudeep George, Vice President – Engineering at iMerit for an exclusive conversation.

Sudeep George is the vice president of engineering at iMerit, where he develops production-ready frameworks for a data-centric approach to machine learning. He has a strong background in imaging sensors, and computer vision and has built and manufactured multi-sensor computational imaging platforms for several market verticals.

 

  1. Could you please brief me on iMerit, its services, and its focus in India?

iMerit is the leader in providing high-quality datasets for machine learning algorithms and artificial intelligence applications. Our offerings include all forms of data labeling and enrichment, including computer vision, natural language processing, transcription, and other content-related data services. iMerit takes unstructured data across different data modalities and annotates them to create highly accurate structured data for advancements in machine learning and artificial intelligence. To ensure the highest quality data is achieved, iMerit provides end-to-end service to guide data scientists through all phases of their data annotation project and architect a solution that meets their needs.

Our company was founded in 2012 by Radha Basu, who led the IPO of Support.com and spent 20 years at Hewlett Packard, where she started HP’s operations in India. iMerit employs more than 5,500 full-time data annotation experts in Bhutan, Europe, India, and the United States, with 80% of employees coming from impact backgrounds and women accounting for 52% of the workforce. iMerit works with leading brands across the globe, with Fortune 500 companies in healthcare, autonomous mobility, geospatial technology, BFSI, retail and e-commerce, agricultural AI, legal, and others.

Today, we’re starting to see AI and ML become an integral part of enterprise digital transformation. As the AI market in India moves closer to production and deployment, iMerit aims to provide the expertise and data intelligence required to scale ML projects.

 

  1. According to you, what differentiates iMerit from other AI Data Solution companies in India?

iMerit provides an unparalleled level of expertise at all phases of the data labeling process. Unlike many other companies, iMerit comprises full-time employees, not contractors or crowdsourced labor. We have over 90% retention and invest heavily in learning and development to build expertise in every AI domain and industry, across tools, the data modalities being annotated, and the data workflows designed. Through our learning and development process and industry-dedicated expert, we have achieved more than 98% accuracy in our data-labeling services.

Solving edge cases, i.e., rare occurrences that an ML algorithm has never seen before, is critical to the success of AI. This is often referred to as the problem of the long tail and is the single biggest challenge in deploying AI solutions to the real world.  iMerit’s data experts are highly skilled at identifying and resolving edge cases.  With iMerit’s edge case platform, clients can gain visibility into edge case resolution, view edge case insights and analytics, and access a repository of edge cases for future projects.

We strongly believe that technology, talent, and technique must come together as an integrated, cohesive offering to enable our customers to achieve their goals. We build in-house technology that allows our talent to utilize the best techniques in delivering value to our customers.

Our top-of-the-line market solutions include a full-time in-house annotation workforce, industry subject matter experts, customized workflow, ISO & SOC2 Type II Compliance, in-house people training, expertise with edge case scenarios, and the iMerit Data Studio, which is a platform that brings together tools that manage configuration, annotation, project progress, and analytics, and more in one end-to-end solution. These have paved the way for us to emerge as a leading AI data solutions provider.

  1. AI has become the driving factor in every industry. How do you think AI has evolved over the last couple of years?

Over the last two years, the AI ecosystem has seen a push to move from a model-centric AI approach to a data-centric AI approach, resulting in improved performance of AI systems.  This shift, championed by AI pioneer Andrew Ng, has had a dramatic impact on the quality of machine learning models deployed. Today, AI enables companies across industries to increase productivity, create unique capabilities, and improve customer experience. CIOs are investing heavily in AI to sustain growth, build a future-ready workforce and stay ahead of the curve. Over the last few years, the initiatives in AI at the policy level established by the Government of India have driven significant interest in AI across sectors. AI in India is now empowering the nation’s workforce and enabling enterprises to access and build innovative solutions and capabilities, thereby strengthening the nation’s democracy. Campaigns like “AI for India” and computing platforms like AIRAWAT propel India toward an AI revolution.

  1. How does iMerit help businesses across multiple industries annotate their data and develop unique AI solutions that drive their company’s growth?

As I mentioned, iMerit’s clients range across various sectors, including autonomous mobility, geospatial technology, healthcare, finance, commerce, government, and agriculture. Our teams label thousands of images, videos, and documents to create a highly accurate ground truth that directly impacts an ML algorithm’s performance.

For example, iMerit’s agriculture AI experts have annotated drone data to train computer vision models to identify objects accurately and automate repetitive tasks like harvesting, picking, seeding, and spraying. We have helped eliminate the risk of crop damage by negligence and allowed farmers to focus more on improving overall production yields. Besides agricultural AI, iMerit also advances drone mapping for military and urban development.

iMerit’s medical experts label text in digital documents, such as clinical trial data, to power robotic process automation, clinical decision support algorithms, and virtual assistants. They have annotated thousands of plain films, CTs, and MRIs to drive innovation in instrument tracking, lesion detection, and phase identification.

Chatbots, voice assistants, recommendation systems, and augmented reality systems are some examples of AI products that are increasingly used in everyday life. These all use data annotation services like text/image annotation, categorization, content moderation, sentiment analysis, and audio transcription to improve accuracy and performance.

 

  1. What role does the Data-Centric AI approach play in building a custom AI system? How does this approach help in navigating today’s complex technological environment?

I briefly mentioned the problem of the long tail. It is a situation where the AI system has to handle situations it is not trained for, and given the complexities of the real world, this quickly becomes a continuous point of failure. The earlier approach to building AI systems was to continuously improve the system performance by focusing on tuning and tweaking the ML model parameters. Unfortunately, this model-centric approach mostly does not help improve real-world performance to the required level. In the data-centric approach, the focus is on training the model with a diverse set of training data that best represents the real-world conditions in which the AI system has to operate. This approach has enabled AI systems to handle complex scenarios much more practically and efficiently.  Generally, an AI system is only as good as its training data. The data-centric approach has enabled AI to be deployed more successfully in various domains.

 

  1. Going forward, what trends do you expect to see in AI in the next five years and how will these trends impact the growth of iMerit’s clients?

More AI experts will prefer synthetic data and unsupervised learning methods in the coming years since the burden of test set data collection and preparation is virtually absent in unsupervised learning. Sensor Fusion and adopting a multi-modal data paradigm will gain more relevance in the coming years. Data from multiple sensors will be gathered and fused to produce more reliable information with less uncertainty.

As we move towards a digital future backed by rapid AI innovations, iMerit’s work and growth will reflect the changing priorities of teams working to leverage ML across industries. Over the next few years, iMerit will make significant investments in technology to better enable AI data solutions across the lifecycle of products and services.

Our focus will be to develop and provide a complete package to clients from a workforce and technology perspective. This will involve an investment in the company’s technology, infrastructure, and ability to scale alongside our clients. Going forward, we will invest in developing solutions to help our customers build complex ML DataOps pipelines, leveraging our expertise in ML, edge case scenarios, anomaly testing, and synthetic data.  Our mission is to enable our customers, spread across various domains and industries, to successfully build and deploy AI solutions.

Source