Web Pages - ScaleDown

Scraped pages and enriched records for GTM and signal products. Four recipes, one per ScaleDown model.

Compress

Sits in front of your LLM. Open the full Python recipe.


Why it fits	Raw web content is mostly navigation, markup, and restatement across sources: high redundancy, high Compress leverage.
How it goes	Scraped page in → Compress strips boilerplate → the downstream model works on the signal.
Payoff	Large token reduction on noisy pages before the expensive step.

Replaces an LLM call. Open the full Python recipe.


Why it fits	The output a user reads should be a tight profile, not a wall of scraped text.
How it goes	Page or record in → Summarize produces the profile → surface it in the product.
Payoff	Readable summaries at crawl volume.

Replaces an LLM call. Open the full Python recipe.


Why it fits	Enrichment is continuous extraction over raw web content; this is the core loop for GTM and signal products.
How it goes	Page in → tell Extract the fields → it returns structured facts → into the record.
Payoff	Structured enrichment off unstructured pages.

Replaces an LLM call. Open the full Python recipe.


Why it fits	Most scraped records are junk; you want a cheap, fast filter before spending on extraction.
How it goes	Record in → Classify scores relevance and assigns tags → route or drop.
Payoff	Fast filtering keeps junk out of the expensive pipeline.