Vijay Vankadaru

Vijay Vankadaru

ML Researcher @ DASION
Graduate Researcher @ UC Berkeley
CTO @ AGMNT

What It Felt Like to Finish a Paper

1 minute read

Published:

The hallucination survey went live on MetaArXiv in March. I want to write about what the process was actually like before the memory fades.

The hardest part wasn’t the writing. It was the argument. A survey paper isn’t just a literature review — it’s a claim about what the literature means. Ours was: hallucination in medical LLMs isn’t a bug, it’s a structural property of the generation process, and the field has been asking the wrong question by trying to eliminate it rather than designing systems that manage it. That argument had to be earned, not just asserted. Every section needed to build toward it.

I rewrote my sections more times than I expected. Not because the facts were wrong, but because the framing kept being slightly off — too focused on what individual papers found rather than what they collectively revealed. Prof. Roosta was patient about this in a way that I found both reassuring and instructive. The feedback was always about the argument, not the execution.

The other thing I didn’t anticipate was how much the writing clarified my own thinking. I thought I understood the hallucination literature reasonably well before we started. I understood it much better after I had tried to explain it to a reader who knew nothing. The act of writing forced precision in a way that reading didn’t. There’s something about having to construct a coherent narrative out of fifty-plus papers that reveals gaps in your understanding you didn’t know were there.

What I’m left with is a paper I’m proud of and a much clearer sense of what I want to work on next. The follow-on work on mitigation methods is already underway. The survey mapped the problem. Now we get to work on part of the solution.

One paper down.

You May Also Enjoy

Why Multi-Instance Learning is Actually Beautiful

2 minute read

Published:

There’s a class of problems in ML that most supervised learning frameworks can’t handle cleanly: you know the label for a group of examples, but not for any individual one.

CTC Alignment and Why Temporal Correspondence Matters in Multimodal Learning

2 minute read

Published:

When you fuse audio and text representations, the obvious approach is to encode both independently and then concatenate or cross-attend. It works. But it misses something important: the correspondence between what was said and how it was said, at the same moment in time.

Industry Experience Is a Research Asset, Not a Gap to Apologize For

1 minute read

Published:

There’s a version of the story I told about myself for a while that went like this: I spent years doing applied work in industry before getting serious about research. That framing treated the industry work as a detour — something to acknowledge and move past.