Artificial Intelligence Brings New Perspective to NJIT Institutional Data Sources
New Jersey Institute of Technology is applying an artificial intelligence layer to its institutional data resources, so that anyone in the NJIT community might find the information they need delivered faster than searching manually and perhaps served with a side of unexpected insights.
The new interface is called IRIS — Institutional Resource Intelligence System — and it’s available now for pilot users. A university-wide rollout is due in the next few weeks.
“The idea came from a practical need that many universities and large organizations face today. Across departments, the same data requests are often repeated, creating duplicate work and slowing down daily operations. At the same time, important information is spread across different systems, dashboards, reports, and policy documents,” explained Xilin Zhang, associate director in NJIT’s Office of Institutional Effectiveness.
“As a result, staff, faculty, and administrators can spend a lot of time searching for information that already exists instead of focusing on their work and decision-making. … I saw an opportunity to apply AI and systems-thinking principles to reduce this operational friction, minimize duplicated work and make institutional knowledge more accessible.”
Users interact with IRIS like any other chatbot — it resides on a private Amazon Web Services cloud but it’s powered by open-source versions of language models from OpenAI, the same company that makes ChatGPT. However, “What sets IRIS apart from a generic AI chatbot is that it was built specifically for the institutional context,” said data analyst Dong Dinh, who built the first version.
“It understands how to navigate different types of university information, knowing when a question is about a data point versus a policy versus a definition, and responds accordingly. It also has built-in checks to flag when it's not confident in an answer rather than making something up. That kind of domain-aware design is what makes it more than just a chatbot with a search bar,” Dinh explained. “The architecture also means we can update the underlying data anytime without retraining, and scale to new sources and users as the platform grows across campus.”
IRIS currently draws from several hundred institutional documents spanning years of data tables, policy handbooks and reports. It also draws from more than 600 metric definitions such as enrollment data, graduation rates and retention information. “The challenge was never about raw file size. Institutional data is dense. A single page of a Common Data Set report can contain hundreds of data points that each need their own context to be interpreted correctly. The real complexity, and where most of the engineering effort went, is in designing a system that knows how to search across all of these different sources and pull back the right piece of information for a given question in seconds,” he said.
In testing, when a user did not specify a date range for requested data, IRIS provided year-over-year comparisons and highlighted trends. Behind the scenes, it runs what administrators called a sanity check every morning, in which the system asks itself more than 100 questions and confirms that it’s still delivering the right answers.
It took a village to transform IRIS from code on Dinh’s computer into a secure system for the entire Highlander community. Three graduate students from NJIT’s Ying Wu College of Computing helped out — Amish Faldu, an AI and machine learning engineer who now works at Adobe; Shambhavi Parasahr, who helped build the user interface; and Deep Talreja, the OIE office’s graduate assistant who did significant work on the cloud deployment. NJIT’s own Information Services & Technology Core Systems team also assisted to merge IRIS onto the university network, Zhang noted.
Looking forward, “Our vision is for IRIS to become a university-wide intelligence platform that helps users quickly find the right institutional resources including data, dashboards, reports and policies. In the future, we see opportunities to expand its role in workflow support, decision support, metadata discovery and potentially broader institutional knowledge management across multiple units,” Zhang observed.
“Our long-term goal is to make trusted institutional knowledge accessible in seconds, enabling faster and more informed decisions across the university.”