|
Listen to this story
|
Former Google and Meta engineer, Debarghya Das, who also goes by the name Deedy – a tech fanatic – recently got in touch with AIM and spoke at length about his journey into the world of AI, alongside busting myths around internal enterprise search and more. He is the founding engineer of internal enterprise search space Glean – an AI-powered workplace search engine that uses deep learning-based language models to provide personalised answers to natural language queries.
Hailing from Kolkata, Das spent most of his childhood back home before heading to Cornell University for his undergrad course and a master’s in computer science. Much like other Indian kids, he was also tutored in FIITJEE, which he found interesting only until competition came into the picture.
Later, he spent a year working for Meta in New York before joining the Google Search team for four years. At Google, from being an intern to a tech lead, he focused on improving the understanding of search queries and later expanded and developed the cricket feature on Search, along with other functions such as Assistant, Google Feed, and notifications. His work required him to work from different locations such as Tel Aviv and Bangalore.
Leaving Big Tech
Before leaving Google in 2019, Deedy was offered to join the planning team of Waymo, Alphabet’s subsidiary of self-driving cars. He shares that although he was thrilled about the job opportunity in his “favourite city, New York”, he felt something was not right and eventually left.
According to him, Google’s slow speed of execution, sluggish professional and technical growth, diluted talent pool, and executives’ aversion to taking risks was a huge deterrent for him. While the company grew and diversified its product offerings, Deedy felt Google had become extremely promotion-centric, not customer-centric.
And then Glean happened. “I was looking for my next opportunity and knew that I wanted to join a smaller venture where the focus was on building a product from zero to one. However, the earlier the stage, the more is the risk. I had faith in the strong and experienced founding team of Glean and that is how I landed up here,” he said.
Founded in 2019 by a former distinguished engineer at Google and co-founder of $5B+ Rubrik, Arvind Jain, California-based Glean hit unicorn status just three years after its launch, led by Sequoia Capital.
Why Not India?
As of now, Deedy has no plans to work or build his own company in India. He enjoys working on core technical challenges and cutting-edge software innovation. In India, most innovation is tech-enabled businesses like education technology, food delivery or e-commerce. The medium may be technology, but the product is still mostly limited to these. It’s important for me as a technical person that the final product not just be software, but cutting-edge softwares,” he said.
Back when big tech chiefs Satya Nadella and Sundar Pichai visited India last year, they echoed similar sentiments on how India is the number one country in terms of AI reports or projects, and that it has the human capital to do innovative things in the context of creating better financial services, risk assessments, insurance, and energy transition.
Deedy, however, is not swayed by the rhetoric. He said “While Indians abroad have undoubtedly been at the forefront of research, innovation and industry leadership, India itself has never been a vanguard of technology. It has the talent, but lacks the funding and resources to be an AI leader.”
Adding to that, bad financial health has plagued many of its unicorns and the typical startup model that involves acquiring a lot of users has not been successful in India due to poor user retention and incentives. “Acquiring users in Tier-2 and 3 markets is key to success in the Indian consumer startup market,” he added.
Google and ARTPARK’s ‘Project Vaani’ will open-source datasets to promote the inclusion of diverse regional and local languages in the Indian government’s Digital India efforts. Meanwhile, Microsoft’s ‘Project Bhashini’ aims to make a variety of Indian languages accessible through translation. But the volume of training data still remains low. “The training data for Indian language speech recognition and synthesis models is insufficient, but the trend towards openness is making more open-source models available,” he stated.
Quest for the Best Enterprise Search
The industry has been unsuccessful in finding a clear winner for enterprise search despite attempts since the 90s with shifts in platform and technology. Building and maintaining enterprise search involves technical challenges of improving ranking with low search volume and high security, as well as non-technical considerations like pricing, stakeholder management, and customer relationships. Glean assists with onboarding, learning, information recall, and remote work.
Convincing customers that your product outperforms competitors is a challenge, especially in a tough economic climate where IT leaders prioritise cost-cutting tools. “Adoption rates are dependent on the rollout strategy and customer relationships. However, they have a retention rate that’s comparable to consumer products like YouTube and Facebook — rare in enterprise SaaS”, he added.
Big Tech’s Big Competition Lies Elsewhere
Consumer search engines are a lot different from enterprise search engines. The latter handle permissioned documents for a limited user base, require a company-specific language understanding, and prioritise security due to low query volume and machine learning model building challenges.
“Big tech isn’t interested in dominating this space since the effort-to-reward ratio is low compared to other areas like cloud products where they have more expertise and less competition. They operate at a different scale, so they’re not considered direct competition. We have a strong partnership with Google Cloud and focus on making customers happy,” Deedy concluded.
