thought it said Large Lagrange Models

Biggest gap I see in most "LLM for practitioners" guides is they skip the evaluation piece. Getting a prompt working on 5 examples is easy — knowing if it actually generalizes across your domain is the hard part. Especially for analysts who are used to statistical rigor, the vibes-based evaluation most LLM tutorials teach feels deeply unsatisfying.

Does this guide cover systematic eval at all?

Totally agree it is critical. Each of chapters 4/5/6 have specific sections demonstrating testing. For structured outputs it goes through an example ground truth and calculating accuracy, demoing an example comparing Haiku 3 vs 4.5.

For Chapter 5 on RAG, it goes through precision/recall (with emphasis typically on recall for RAG systems).

For Chapter 6, I show a demo of LLM as a judge (using structured outputs to have specific errors it looks for) to evaluate a more fuzzy objective (writing a report based on table output).

I’m always curious why local models aren’t being pushed more for certain types of data the person is handling. Data leakage to a 3rd party LLM is top on my list of concerns.

I am not as concerned with that with API usage as I am with the GUI tools.

Most of the day gig is structured extraction and agents, which the foundation LLMs are much better than any of the small models. (And I would not be able to provision necessary compute for large models given our throughput.)

I do have on the ToDo list though evaluating Textract vs the smaller OCR models (in the book I show using docling, their are others though, like the newer GLM-OCR). Our spend for that on AWS is large enough and they are small enough for me to be able to spin up resources sufficient to meet our demand.

Part of the reason the book goes through examples with AWS/Google (in additiona to OpenAI/Anthropic) is that I suspect many individuals will be stuck with the cloud provider that their org uses out of the box. So I wanted to have as wide of coverage as possible for those folks.

but they claim your data is private and they will totally not share any of it with their advertising partners!

Worth noting that AWS Bedrock makes it easy to have zero retention with premier claude models. Not quite local, but it feels local-adjacent for security while getting affordable access to top-performing models... GCP appears to be a bit harder to set this up.

IMO Google Vertex is not any harder than AWS. AWS biggest pain is figuring out IAM roles for some of the services (batching and S3 Vectors -- I actually cut out Knowledge Bases in the book because it was too complicated and expensive). Have not personally had as big an issue figuring out Vertex.

I do have a follow up post planned on some reliability issues with the APIs I uncovered with compiling the book so much -- I would not use Google Maps grounding in production!

I want to buy this book! But the price is too high. Can you offer some kind of HN discount?

You can use `LLMDEVS` for 50% off of epub (that was the coupon I sent to folks on my newsletter).

purchased. thank you!

Is "CRIME" an acronym?

Or is this actually a law enforcement related example?

Click the home button at the top left or just chop off the url. Here: https://crimede-coder.com/

Right there it says in a big page width box

  CRIME De-Coder
  Customized Consulting services, focused on crime analysis for police agencies.
  Contact Me

Crime De-coder is my consulting firm (not an acronym), but the book is not specific to crime analysis -- it is more general.

Would that book be useful as a reference to introduce data journalism students to AI? I'm less interested in the basics of using the API or claude code etc than best practices for workflows dealing with unstructured data, entity extraction, automated pipelines (with evals)? Although I do have some decent workflows around this I'd be interested in reading from someone who lives and breathes this kind of work. Pure data analysis to me is also something where I haven't found a good bridge between the current "generate a python script for me that I'll double check" paradigm and the spreadsheet centric world of most data journalists.

The book is likely a good fit to this type of work. The chapter on structured outputs shows how to extract out data from text, walking through prompt engineering and k-shot examples to generate json, to pydantic, then batch processing with the different providers.

It also shows how to set up evals in different parts of the book. (Depending on what you want to do, the structured outputs has evals show comparing models/prompt changes to ground truth, and the agent chapter has evals LLM as a judge.)