welcome, quick question

Hey,

I’m glad you’re here and enjoyed a public blog post or talk I did.

I write about building real products and workflows on documents. Most of that comes down to turning difficult PDFs into useful, trustworthy data: OCR, structure, retrieval, citations, evals. I want to share useful observations as I find them all in one place, and I want you to be able to reply (that's what this newsletter is).

I read every response. Sometimes those replies turn into the most useful discussions I have all week, and they help me write better posts.

There won’t be a fixed schedule. I’ll email when I have something worth sending.

Most emails will be short: a useful idea, tool, lesson, mistake, or question. Sometimes they’ll go deeper.

I’ll also sometimes tell you about things I’m building, selling, or looking for feedback on. No weird pretending otherwise. But there’ll never be an email that provides no free value and information.

If this ever stops being useful, there’s an unsubscribe link at the bottom of every email.

For now, hit reply and tell me two things:

1. What are you working on right now?

2. What documents does your product need to understand, and where is the workflow letting you down?

Talk soon,

Isaac

P.S. Adding this address to your contacts is the best way to help future emails land in your primary inbox.

P.P.S. Over the next few emails, I’ll send the pieces I’d point you to first: how to choose an OCR model, pipelines versus document VLMs, citations, two kinds of evals, and the document problems that remain after those choices. I’ll also show you the private testing setup behind the work. After that, no fixed cadence.