
Pathway participated in the 6th edition of the Kharagpur Data Science Hackathon, organized by the Kharagpur Data Analytics Group.
Pathway participated in the 6th edition of the Kharagpur Data Science Hackathon, organized by the Kharagpur Data Analytics Group.
This was a chance to work at the actual frontier of AI. Teams went hands-on with Pathway's open Dragon Hatchling (BDH) repository: a starter implementation of our post-transformer architecture, small enough to run on toy datasets but real enough to take apart. Rather than chase a leaderboard, participants ran the model, probed how it behaves, and explained its interpretability properties: how sparse its activations are, which neurons fire, and what they appear to specialize in. It meant getting inside one of 2025's most discussed AI papers, reasoning about an architecture modeled on the brain, and building the kind of research instinct that turns a strong engineer into a researcher. The repository exposes published components only, a window into the architecture rather than the full system, so the work was about what BDH reveals, not reproducing capabilities such as continual learning or infinite-context reasoning, which are not part of the public repo.

