Modern medicine is based on data organized in tables and graphs to help doctors make quick, informed decisions. Survival rates and prescription side effects are often extracted into a format that can be scanned in seconds. But that clarity doesn’t come easily. Data must be extracted and organized, an often time-consuming process that can leave important information buried deep within research reports.
Naman Ahuja wants to change that.
This May, Ahuja will graduate from Arizona State University’s Department of Computing and Augmented Intelligence, part of the Ira A. Fulton College of Engineering, with a master’s degree in computer science.
Over the past two years, his research has focused on how to obtain artificial intelligence (AI) systems to transform long, unstructured texts into accurate, usable tables. This spring, his work earned him the IBM Infrastructure Masters Fellowship Award. This prestigious award recognizes research that has had a strong impact on the real world and industry.
Looks right, looks wrong
This issue reveals an important limitation of modern AI. Large language models make it easy to read and summarize documents. But when asked to extract accurate information and organize it into something structured, like a table that doctors and analysts can rely on, they often struggle. Important details may be missed, information may be inconsistent, or the model may generate claims that are not supported by the original text.
Although the results seem elegant, they are not always reliable. Ahuja’s research focuses on filling that gap.
“In the real world, a lot of data exists in complex, semi-structured formats, such as PDF documents and Wikipedia pages,” he says. “While these documents have some structure, they are still complex and contain a lot of information.”
His solution, developed through a master’s thesis, rethinks how AI should approach problems.
Rather than asking the model to generate a table in one pass, Ahuja breaks the task into steps. First, the system extracts atomic facts from the text. Next, make a plan for how you want to organize your table. Finally, populate the table step by step, updating entries as new information appears.
This approach mirrors the way humans perform the same tasks. That is, read carefully, decide which categories are important, and fill out the table bit by bit. His paper argues that reliable structured generation is about breaking down complex tasks into smaller verifiable steps, reducing errors and improving traceability.
In high-stakes environments like healthcare, traceability is paramount. There, clinicians often conduct systematic reviews, reading large amounts of research and distilling important results into tables for decision-making. This is a time-consuming manual process, and mistakes can have serious consequences.
“By taking these disparate documents and converting them into a structured data format, you can more easily access that information,” says Ahuja. “This reduces repetitive data extraction and allows clinicians to focus more on interpreting results.”
Ahuja’s approach is also designed for what researchers call living data: information that evolves over time. As new research is published, systems like the one he is developing can update existing tables rather than rebuilding them from scratch, incorporating new evidence while remaining consistent.
The focus on reliability and practical ease of use is one of the reasons IBM’s fellowship program attracted so much attention.
Vivek Gupta, assistant professor of computer science and engineering at the Fulton School and director of ASU’s Complex Data Analysis and Reasoning Lab (CoRAL), where Ahuja conducted research, sees this work as part of a broader shift in AI.
“Naman’s work is a great representation of what we are trying to do at CoRAL. We focus on complex structured data, and specifically how to generate and properly evaluate it, so that people can build trustworthy AI systems in real-world environments,” says Gupta. “He was very thoughtful about both his methods and how he evaluated them. The IBM fellowship was well-deserved.”
turn the tables on the future
For Ahuja, the path to that job began in Hyderabad, India, where she earned a bachelor’s degree in computer science and came to ASU in 2024. His focus has always been on transforming research into something that can be used beyond the lab.
At ASU, he served as a teaching assistant for a graduate-level natural language processing course, gave guest lectures on neural networks, and presented research at an international conference in Vienna. Along the way, he credits Mr. Gupta’s guidance with helping him overcome the inevitable setbacks of research.
“Dr.Gupta guided me through all the different difficult aspects, which is important when an experiment doesn’t work out the way you thought it would,” says Ahuja.
Outside the lab, Ahuja tries to maintain balance. He regularly plays basketball, explores new music, and relaxes by watching stand-up comedy in Hindi and English.
As he prepares to graduate, Ahuja has already stepped into the next phase of his career, accepting a full-time role at Amazon in Seattle, where he will continue to work on large-scale systems development.
“I’m interested in core systems, how these models are actually built and how they can be used to solve real-world problems,” Ahuja says.
The move aligns closely with his long-standing interest in applying research to industry challenges. And while his immediate future is already decided, the broader issues he focuses on remain unresolved. The world is producing more text than ever before, and the need to turn that text into usable knowledge is only increasing.
For now, that still means someone somewhere is building tables by hand. Ahuja’s research suggests that this is not necessary.
