A loan recommendation, fraud alert, or supply-chain forecast can be useful to a company until someone asks why the system reached that conclusion. Siddhant Raman has built much of his work around what happens next. As a software engineer focused on cloud systems and artificial intelligence, he designs technology for large enterprises while also concentrating on whether those systems can explain the reasoning behind the recommendations they produce.
“The question cannot come after deployment,” Raman says. “If a company may eventually need to explain why an AI system produced a certain recommendation, that has to influence the way the system is designed from the beginning.”
Raman sees that issue becoming more urgent as artificial intelligence moves beyond simple assistance and toward greater autonomy. Agentic AI systems can plan and execute multi-step tasks with less direct human involvement, creating possibilities for automation across areas such as supply chains, inventory management, and enterprise operations. Raman believes those systems will become increasingly important, but he is skeptical of how quickly some companies are handing over decision-making authority. His concern is not that autonomous systems have no value. It is that an incorrect early assumption can affect later steps when an AI agent is allowed to continue acting with limited oversight.
“A wrong assumption early in the process can affect every step that follows,” Raman says. “That is why greater autonomy has to come with safeguards.”
That is why Raman favors human-in-the-loop approaches for high-stakes enterprise decisions. He sees AI as particularly effective at processing information, identifying patterns, and producing recommendations. He does not believe those strengths remove the need for a person to make the final judgment when a decision carries financial, legal, or operational consequences.
Raman’s view is grounded in enterprise engineering work rather than theory alone. At Informatica, he worked on AI-driven data integration and cloud systems, and he now works at Oracle designing cloud-native distributed services for enterprise applications. Both roles have involved enterprise systems designed for large-scale use. They also reinforced his belief that an AI recommendation is only part of the problem. Businesses need to understand how that recommendation was produced and whether they can justify acting on it.
“Accuracy matters, but it is not the only question,” Raman says. “A company also has to ask whether it can trace the decision and explain the reasoning behind it when someone challenges the result.”
Raman has developed that argument through technical writing as well as engineering. His work has appeared in Enterprise Tech Journal, Cloud Computing Magazine, and IEEE Computer Society publications, with his published work receiving more than 200 citations from researchers and practitioners. He has also been invited to judge hackathons at UC Berkeley, UCLA, UC Davis, UC Santa Cruz, and Code4Hope. Those roles place him in a different position from his day-to-day engineering work, asking him to assess how other builders approach technical problems and develop new applications.
For Raman, the larger issue is how companies think about trust. He believes many organizations still approach explainability as something added after a model is built, often when legal, regulatory, or reputational questions appear. His position is that transparency works better when engineers treat it as part of the architecture rather than as documentation created later.
“You cannot bolt explainability onto a system at the end and expect it to be meaningful,” Raman says. “The ability to understand how a decision was reached has to be considered while you are building the system itself.”
That position shapes the way he views artificial intelligence in heavily regulated industries such as finance and healthcare. Organizations in those sectors may rely on AI-supported recommendations, but responsibility for the resulting decisions does not automatically transfer to the software. Raman believes that distinction becomes more important as algorithmic systems take on a larger role in how companies evaluate information and choose what to do next.
Raman argues that expectations about AI often become unrealistic when people assume human judgment will simply disappear. Artificial intelligence can analyze large amounts of information quickly and identify patterns that might otherwise be missed, but it does not assume accountability for the consequences of a decision.
“AI can improve the information behind a decision without becoming the person responsible for making it,” Raman says. “There still has to be someone who understands the context and owns the final call.”
Raman believes the strongest applications will combine AI capabilities with explicit human oversight rather than removing people from the process entirely. In his view, the goal is to increase what AI can contribute without weakening the accountability that remains with the organization using it.
He also plans to carry that emphasis on trustworthy AI into a future entrepreneurial venture, which he says will focus on making sophisticated enterprise AI more accessible to mid-sized companies. The idea remains part of his longer-term direction rather than an established product today.
For Raman, the standard for enterprise AI should rise as the systems gain more influence. A recommendation that affects a consequential decision should come with enough transparency for the organization using it to understand what happened and determine where human responsibility begins.
“A system becomes more useful when people can understand why it reached a decision and know where human judgment still belongs,” Raman says. “That is what makes trust something you can design for rather than simply ask for.”