{"projects":[{"slug":"proof-of-personhood","title":"Proof of Personhood at Scale","blurb":"Machine-learning systems that help verify a unique human without surveilling them, privacy-preserving models running at planetary scale.","tags":["ML","Privacy","Biometrics"],"year":"2024 →","status":"Tools for Humanity","url":"https://eklavyagoyal.com/work/proof-of-personhood","problem":"The internet is about to be flooded with agents that are indistinguishable from people. Almost everything online - accounts, votes, benefits, attention - assumes one human behind one identity. Proving uniqueness usually means surveillance: collect everything, compare everything. The problem is doing it the other way around - verifying that someone is one real human while learning as little about them as possible.","approach":"Machine learning under physics-grade constraints. Models whose failure modes are characterized before they ship, because at this scale 'mostly right' is wrong for millions of people. Privacy is a design input, not a compliance step: the systems are built to verify uniqueness without building a profile of the person being verified.","result":"Production ML inside World's verification stack - learning systems doing real work, for real people, where the cost of being wrong is measured in trust rather than benchmarks.","stack":["Applied ML","Computer vision","Privacy engineering"],"links":{"live":"https://world.org"}},{"slug":"physics-informed-learning","title":"Physics-Informed Learning","blurb":"Neural networks with the laws of physics baked into the loss, models that respect conservation laws instead of fighting them.","tags":["PINNs","Simulation","Research"],"year":"2024","status":"TUM","url":null}],"source":"https://eklavyagoyal.com/work"}