I'm Aditya Bansal, a CS undergrad at BITS Pilani, Dubai Campus who got interested in one question — how can you trust a model you can't fully see inside?
That question pulled me towards adversarial ML and AI safety. Over the past year I've been working on backdoor detection in LLMs, adversarial attacks on vision-language models, and algorithmic auditing of real consumer applications. The thread across all of it is the same: understanding how existing systems fail, and eventually designing safer ones.
I'm currently building towards predoc and research fellowship roles before a PhD in adversarial ML / AI safety. I've just chosen a path, and I'm figuring it out on the way.