I built a matcher meant to finish the cleanup that normalization left behind. Testing it against real data showed that no version of it could be made safe. What follows is the architecture that was left once the matcher was set aside.
Enterprise Document Intelligence [Vol.1 #B3] - A confident wrong answer is a bug. A bare âno answerâ with no justification is almost as bad. Each of the four bricks has one piece of evidence to show
A visual guide to how graph neural networks work under the hood
Why the standard groupBy function isnât enough
Your JSON Is Valid but Your Data Is Wrong: Five Failure Modes LLM Structured Outputs Won't Catch
Large Language ModelsFive failure modes that survive constrained decoding, and why your schema validator will never catch them.
The hidden assumptions behind the data we observe.
More informed decision-making through uncertainty quantification
What each one solves, and runnable code you can paste into a notebook.
What I learned after thinking more carefully about Structured Outputs on messy, incomplete data
