In April we started preparing proposals for two funding programmes: the Google TPU Research and Education Awards 2026, and the Amazon Research Awards, Spring 2026 cycle. If selected, both come with real infrastructure: TPU access and mentorship from Google's engineering teams on one side, AWS Trainium credits and a dedicated Amazon research contact on the other.
The plan was for both to sit below the framework, where you think about memory hierarchies, compiler behaviour and how a model actually moves through silicon. That is the layer that changes what it costs to train and serve large models, and it is hard to do alone.
The call
So I posted a call for one or two collaborators, with first preference to MJCET students in the 2027 and 2028 batches. The requirement was depth, not a publication record: JAX, custom kernels, systems thinking or low-level optimisation. The application was a GitHub link, a project, or anything that showed how the person thinks.
By July, four proposals were under review, all with Dr. Fahmina Taranum as principal investigator. What each one asks is in Four proposals, one question each.