Eric Bae

I've been taking things apart since I was a kid — toys first, to see the inner machinations, and eventually the algorithms behind them. That same itch is what pulled me into research. These days I'm a CS undergrad at the University of Washington, and I spend most of my time in the RAIVN Lab on offline reinforcement learning — specifically, how the way we report results quietly shapes the conclusions we draw from a benchmark. Our current work trains 145,000+ policies across 114 datasets to pull that apart.

Before that I designed a Sequential VAE for a generative TSP model, classified T-cell DNA sequences at Fred Hutch, and led a TA team building grading and testing tooling for 600+ students a quarter. What ties it all together is the part I actually love: building the thing, measuring it honestly, and leaving behind infrastructure other people can reproduce and build on.