Experimental Studio for Platform Utility & Discovery.
An academic research collective focused on AI and Edge Computing.
…and the training-time loop behind it. A language model can get dramatically better answers by thinking longer at answer time — not only by being retrained or made larger.
Shashank Bangalore Lakshman, Scott Eiers. The Story of Inference-Time Scaling. 9th Workshop on Visualization for AI Explainability (VISxAI), IEEE VIS 2026. Accepted; to be presented.
Why language models make things up, and what to do about it. Fourteen short visual pages on what hallucinations are and why they are a structural consequence of how these models work.
Shashank Bangalore Lakshman, Shravan Joshi. The Confident Machine. 9th Workshop on Visualization for AI Explainability (VISxAI), IEEE VIS 2026. Accepted; to be presented.
A research hub for the Principal Investigator — projects, team coordination and academic deadlines in one local-first web app. No backend, no accounts, no telemetry; nothing you type leaves your browser.
Open-source, MIT license.
Scam detection that never leaves your phone — a 4-bit quantized Gemma running on the iPhone GPU via MLX-Swift. Nothing uploaded, nothing logged, ~2.5s median verdict on an iPhone 15 Pro.
Open-source, MIT license. Submitted to the Gemma-4-Good Kaggle Competition.