BUAI SOCIETY
This year's projects, run by students for students. Pick a group and dive in.
Sign Up for a Project
We have three robotic cars, one robotic spider, and a robotic arm. We'll be working to turn these into a project. The project is still unclear, but we will figure it out.
Denali and Achyut competed in the PXR Datathon hosted by OpenADMET, a challenge to predict a drug's metabolism by predicting its binding to PXR, a key xenobiotic receptor. The follow-up challenge, also hosted by OpenADMET, focuses on predicting binding to cytochrome P450. The enzyme that metabolizes most drugs flagged by PXR. What stood out to them was the community: top teams openly sharing how they reached their results, recurring webinars, and ongoing discussions around the fast-moving field of computational biology.
We previously trained a GPT-2-level model from scratch. Building on that, we're now exploring local LLMs. Frontier labs threaten to monopolize the intelligence space, but we believe smaller on-device models can rival current state-of-the-art performance. We're leveraging new research in diffusion, flow matching, quantization, and custom inference kernels to maximize tokens per second without sacrificing intelligence.