Stevens Master of Data Science graduate Vandna Rajpal and her team, Neural Navigators, received first place at the D3CODE 2025 Global Hackathon. Her team designed Predicting and Reporting Insights with Scalable Models (PRISM), which is an AI-powered data intelligence platform. PRISM allows users to ask questions about their data and receive reliable and secure answers. In an interview with The Stute, Rajpal summarized, “It is like having a data scientist, analyst, and AI assistant rolled into one platform — accessible to anyone, instantly.”
According to Rajpal, PRISM allows users to ask questions about their data in “natural language” rather than having to use SQL or Python. It also automates ETL (Extract, Transform, Load), builds predictive models without requiring coding experience from the user, and understands documents (files, reports, texts, etc.). Additionally, PRISM has built-in governance and ethics, meaning that it “tracks data usage, decisions, and ensures transparency and responsible AI.”
Rajpal explained that PRISM is special for a few reasons, such as decreasing time to insight “from months to minutes,” reducing dependence on data engineers and data scientists, and it is “enterprise-ready” because it is “secure, auditable, and scalable.” Furthermore, PRISM bridges the gap between research and “real-world use.” Rajpal referenced “the ‘last mile’ problem in AI.” The“last mile” problem refers to “the gap between laboratory perfection and real-world implementation.” Essentially, for AI the “last mile” is “the gap between what AI can produce and what an organization can responsibly release into production.” PRISM stands out because of its ability to bridge this gap.
As for the challenges of this project, Rajpal described, “The hardest part was making natural-language AI reliable at enterprise scale. Translating user questions into correct queries and models while enforcing governance, explainability, and access control was extremely challenging. We had to ensure the system was powerful without being opaque or unsafe.” Still, the most rewarding part of the project for Rajpal was “seeing non-technical users get accurate, explainable insights in minutes instead of months. That’s when we knew we’d successfully bridged advanced AI research and real-world decision-making.”
Rajpal’s experiences as a data science graduate student at Stevens followed her to the D3CODE Global Hackathon. Through her coursework, Rajpal became more comfortable working with “messy, large-scale data under real constraints.” She specifically cited the course BIA 678 (Big Data Technologies), taught by Professor Venu Guntupalli. This course “pushed me to actually build distributed pipelines,” which was helpful for the hackathon, “because PRISM isn’t just a model; it’s a system. I was already used to thinking about data ingestion, scalability, and what breaks when data grows or schemas change. That mindset made it much easier to design something practical under pressure.”
Following this experience and her education at Stevens, Rajpal advises students interested in data science to focus on understanding the data and the problems being solved rather than just focusing on coding or learning tools. She urges, “Build projects, experiment, and don’t be afraid to make mistakes, because that’s how you learn what works at scale. Also, work on integrating different skills like cleaning messy data, building models, and making sense of the results because that’s what actually makes you effective in the real world.”
For Rajpal, this hackathon was bigger than just winning. As she says, “Some wins are just personal, and this is one of them.”
