Vaibhav Hariram

CS and Geography at Berkeley. I spend most of my time teaching AI systems to know when they're wrong. Turn out that's the hard part, not the being-smart part.

Right now I'm at Intel doing AI and ML engineering on PMC & Firmware Validation, building Shana: immigration guidance for F-1/OPT/H-1B holders that doesn't require a lawyer just to understand the first form, and doing research with Berkeley RISE Lab's Gorilla Project on function-calling evals. If a benchmark's ever told you your model's tool-use is bad, that might be part of my work.

Before this: Paprika, a runtime policy and replay system for AI agents, running with 25+ teams now. Some infra work at Flywheel. A few PRs into vLLM and PyTorch, not all of them made it — that's fine. Started out doing GIS software work at Railinc, which is where the map obsession actually comes from.

Outside of that- I collect old maps and jerseys, sketch when I'm patient, take too many photos of skylines, and watch Chelsea find new ways to disappoint me every week. Also deep into all types of music, k-dramas and anime.

Talk to me about any of it, anytime.