Neurips 2025 San Diego
Neurips in San Diego this year was amazing. I got to meet brilliant researchers from all over the world in many forms of machine learning!
Enginuity — Dataset Proposal
Stemming from my work at the SNS facility at Oak Ridge, I’m excited to share that my team and I won the Dataset Proposal Competition at the AI4Science Workshop at NeurIPS this year! 🎉 This was also my first talk at a major conference, and I had the opportunity to present a poster to a packed room of researchers facing similar challenges in scientific and engineering data. Our proposal includes a call for community participation — a collaborative effort to help unlock high-value scientific datasets that are traditionally kept proprietary. It was incredibly rewarding to see how strongly the community resonated with the vision behind Enginuity and the broader problem it aims to solve.
My Talk
Unfortunately, my eyes look a little closed in this photo, but I’m still incredibly proud to have presented to such a large audience. It was surreal standing on the NeurIPS stage and sharing work that grew out of months of research, experimentation, and cross-team collaboration. The room’s energy and interest made the entire experience unforgettable.
The Poster
Our poster was a massive hit, drawing a constant stream of researchers and industry folks dealing with similar retrieval and documentation problems. I’m deeply grateful to have worked alongside an amazing team, my mentor Dr. Ghosal, Prahitha, Harshita, Tilak, and everyone from the SNS facility who supported the vision from day one. Talking to attendees throughout the session highlighted just how widespread these challenges are across scientific and engineering disciplines.
The poster showcased the Enginuity initiative: a large-scale effort to build an open, richly annotated repository of engineering and scientific diagrams. We highlighted our phased data-collection strategy, combining public-domain automotive diagrams with an industry-engagement pipeline that enables contributions without exposing proprietary IP. The response was overwhelming — researchers immediately recognized the gap Enginuity fills and were eager to collaborate, contribute diagrams, or integrate the benchmark into their workflows. Many attendees also shared examples of similar problems in their own organizations, reinforcing that this dataset could have a real, long-term community impact. The enthusiasm reaffirmed our belief that Enginuity can become a community-driven cornerstone dataset for multimodal reasoning in scientific and engineering domains.
Paper Preview
If you’re interested in reading the full paper or other related works, please reach out — and keep an eye on my Google Scholar for updates! Below is a preview of the introduction. We’re actively expanding this research, and there will be several updates in the coming months as we refine the benchmark, improve the retrieval pipeline, and prepare public releases.
Sightseeing
San Diego wasn’t all work! It’s a beautiful city, and I had the chance to explore sights like a pirate ship and the USS Midway. The waterfront was incredible, and getting a break from the conference bustle was a perfect way to recharge between sessions. I can’t wait to go back!
