DataStorage.com sits down with the founder reimagining how pharma, and eventually the enterprise, approaches AI.
In the latest episode of our executive webcast series, we sat down with Abhishek Jha, founder and CEO of Elucidata, to unpack the company’s transition into a full-fledged AI player and its vision for the future of enterprise intelligence. Interviewed by DataStorage.com’s own John Kosturos, the conversation spanned rebranding strategy, data governance, GPU utilization, and AI model economics.
Founded in 2015, Elucidata began as a life sciences data platform. But in 2024, the company made a pivotal shift—rebranding around artificial intelligence and positioning itself as a leader in out-of-distribution (OOD) problem detection. Jha argues that the biggest opportunities in AI aren’t about pattern recognition—they’re about identifying and solving edge cases where traditional models break down.
“Any valuable biotech company out there, right, that is based on one rare discovery, a very unique insight, right? A biological insight about a group of patients not responding to a drug or about a group of, you know, humans that have a disease, right? There’s always some unusual thing, right, which in some ways does not match a pattern,” Jha explained. “And the traditional AI tools stack that has been developed has been very, very good at matching patterns.”
Jha shared that Elucidata didn’t jump on the AI hype early on—not even in their seed round.
“Our investors have been great partners for us. They were insisting on using the word AI or the phrase AI, in that press release,” he recalled. “And if you look at that press release, I did not use that.”
In the wake of large language models (LLMs) and growing market validation, Elucidata now leads with a “data-centric AI” approach—emphasizing quality, context, and compliance. Today, the platform supports pharma R&D teams by helping them transform unstructured and multimodal data (text, tabular, imaging) into AI-ready products tailored to specific use cases.
While the interview touched on observability, secure deployments, and customer-managed clouds, the standout philosophy was consistent: build from the problem back. Start with a use case, curate a relevant dataset, build high-quality reusable data products, then repeat.
Jha described a common trap: attempting to make “all data” AI-ready in one sweeping effort, only to find the ROI doesn’t hold.
“We took an approach where that was, I think, a mile wide and inch deep,” Jha said, “where we said, like a lot of other organizations did at the time, that we’ll take all the data that is out there, we’ll clean it all up, get it all AI ready… it was just not giving a good enough ROI for our customers.”
Elucidata’s commitment to precision extends to infrastructure. With growing compute requirements and increasing GPU usage, the engineering team has begun leveraging hybrid cloud setups and bare-metal clusters to manage cost and performance.
Jha also highlighted deployment flexibility: with strong capabilities in structuring and governing sensitive data, Elucidata enables customers to deploy AI tools across Google Cloud (GCP), AWS, and private clouds—via dashboards, APIs, or direct model integration.
Despite the industry trend, Elucidata isn’t building its own foundation model today.
“We haven’t been able to convince ourselves yet that a universal all-purpose foundation model in drug discovery is needed,” Jha said bluntly. “We don’t think the state of the art is there yet.”
Still, he acknowledged that this stance could change as the technology matures.
Although Elucidata’s current focus is on pharma and diagnostics, Jha hinted at broader ambitions. OOD problems aren’t unique to healthcare—they also show up in hedge funds, manufacturing, autonomous vehicles, and anywhere pattern recognition fails.
“We’ve cut our teeth in one of the hardest verticals,” he said. “And we’re ready to take what we’ve learned into others.”
With clients already running dozens of daily queries on Elucidata’s platform, the team is gaining real traction.
“It’s not working perfectly,” Jha admitted. “But it’s a real step in the right direction.”
The takeaway: the future of enterprise intelligence may not be built on models alone, but on the quality and structure of the data that feeds them—plus the operational discipline required to deploy AI responsibly.
