VIDUSHI ANAND
projects
🛡️ SafeToSay
The model wasn’t unsafe, it was unsupervised. SafeToSay just gave it rules to play by.
🏆 FirstPlaceWins
When models aren’t sure, they don’t just guess, they lean toward what comes first. FirstPlaceWins shows that ambiguity quietly turns structure into a decision-maker.
🧩 Build-LLM-Playground
A scratch-built LLM playground inspired by *Build a Large Language Model (From Scratch)*, tokenization, transformers, and tiny GPTs built piece by piece to understand how they really work.
🧪 Difflo
Teaching a UNet to grow flowers from pure noise 🌸, less about perfect petals, more about learning diffusion the hard way.