Building AI We Can Trust: Santosh on Reliability, Research, and What Comes Next

From data engineering to AI research and entrepreneurship, Santosh is focused on one question: How can we build AI systems people can actually trust?

For Santosh, the question behind trustworthy AI started with something simple: the same input should not produce a different answer every time.

His career began in data engineering, building systems where consistency was expected, not optional. But as generative AI rapidly entered more areas of business, he noticed a different standard emerging. The same document could be processed twice

by a large language model and produce two different answers. 

That tension eventually pulled Santosh from large-scale data systems into AI research and entrepreneurship, where his work now focuses on a bigger question: How can we use the capabilities of AI without giving up reliability, accountability, and control?

Making AI More Predictable

For Santosh, trustworthy AI starts with separating two things: making a decision and explaining it.

Instead of asking an AI model to do both, his approach uses deterministic, rule-based logic to make the decision first. AI can then do what it does well. It’s to turning that decision into an explanation people can easily understand. 

His research has put that idea to the test. In one study, 15 documents were each processed 20 times. While a traditional AI model produced different answers across repeated runs, Santosh’s hybrid approach returned the same result in all 300 executions. 

For Santosh, that consistency is what separates a system that can simply generate an answer from one that can be traced, audited, and trusted.

Turning Research Into Something Real

That philosophy became the foundation for TriageCounsel, where Santosh applies his research to contract risk review.

The idea is captured in one simple principle: “Deterministic Rules Decide. AI Explains.”

Rather than allowing AI to decide whether a contract clause is risky, TriageCounsel uses predefined rules to identify and determine risk first. AI comes afterward, helping explain those findings in clear language without changing the original decision. 

In areas like legal services, Santosh believes that distinction matters. A system used for important decisions needs more than a convincing answer. It needs a result people can reproduce and understand.

“Determinism gives you an audit trail; explanation gives you understanding. You need

both.”

From Researcher to Builder

Building the technology, however, has only been part of the challenge.

Santosh says one of his biggest lessons has been learning the difference between creating rigorous research and creating something people will actually use. Developing TriageCounsel has required him to think beyond the technical work, from positioning the product clearly to finding the right collaborators and making the technology accessible to legal teams. 

“Research rewards rigor; building a product rewards clarity.”

It is a balance he continues to carry as both a researcher and entrepreneur: maintaining the strength of the research while making its value clear in the real world.

Turning Technical Skills Into Meaningful Work

For young engineers, researchers, and aspiring founders, Santosh’s advice is not to treat technical ability as the final destination.

“Don’t treat your technical skill as the finish line—treat it as the raw material.”

He believes meaningful work happens when strong technical skills meet a real problem that still needs a better solution. Whether that happens inside an established company or through entrepreneurship matters less than finding that intersection.

Looking ahead, Santosh hopes to explore how the same approach to reliable AI can extend beyond contract analysis. His research raises a broader question about where AI should be allowed to influence decisions, and where results need to remain independently verifiable. 

For Santosh, the goal is not to make generative AI less capable. It is to build clearer boundaries around what we trust it to decide.

Editor: Michaeline stephanie