"Squirrel team on AI Infra Summit Hackathon Hackathon"

Team Idea

Making AI text generation 2-3x faster with zero quality loss. Two models working together: a small one guesses the next few tokens, a big one verifies them in a single pass. Our twist: a live controller that adjusts how many tokens to guess based on what's being written. Code = guess a lot. Reasoning = barely guess at all.

Team Members

marwa kiani

marwa_kiani17

Asjad Sohail

Asjad_Sohail

Submission

AI Crossover point

Make a 7B AI model answer faster by having a tiny model guess ahead — same output, measured up to 2.69x speedup on consumer GPU.