Add Prior Labs agent cookbook

[kacperlukawski]

Adds a new cookbook recipe demonstrating how to combine LLMs and classical tabular ML to enrich scraped web data.

The notebook builds a Haystack pipeline that:

1. Scrapes hotel listings from the web using Firecrawl.
2. Uses Claude to extract and structure the data into a Markdown table.
3. Connects to Prior Labs' TabPFN via MCP to predict missing attributes with predicted values marked in bold.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[bilgeyucel]

Line #13.            # We need to rewrite the previous assistant replies to user

Maybe a better explanation is "this particular provider expects the conversation to start with a user message".

Reply via ReviewNB

[kacperlukawski]

Changed it

[bilgeyucel]

I'm confused here. Don't all conversations end with an assistant message?

[kacperlukawski]

Yes, they do! In our case, the last message from the "summarizer" is an assistant message (a Markdown table with the hotel details), so the conversation ends with an assistant message when we pass it to the agent (that's what Anthropic does not support - the last message has to be a user message with an instruction on what to do next).

But I feel I got the confusion here. A cleaner approach might be to just add a simple user message to the sequence and avoid having this prompt builder. Let me try improve it.

[bilgeyucel]

More of a detailed ask: can you display the content in markdown format, but just the table, for comparison? Splitting the string from "| Hotel Name|" could be beneficial here.

Reply via ReviewNB

[bilgeyucel]

Thanks @kacperlukawski, I love the use case! I left some comments, mostly around the formatting.
Additionally, can you add the notebook details in index.toml, otherwise, it wont be listed on the website

[kacperlukawski]

@bilgeyucel I would appreciate another look at the cookbook

[bilgeyucel]

I only added one extra comment to one thread. Other than that, looks very good!