Routing Engine

Intelligent extraction method selection that picks the cheapest path to quality data.

Pipeline

Extraction Method Selection

The tool router evaluates each URL and selects the optimal extraction strategy:

1. Cache check — If a recent extraction exists (within TTL), return it immediately.
2. Schema.org probe — Fetch the page and check for JSON-LD / Microdata. If found, extract structured data (cheapest method).
3. LLM extraction — If Schema.org is absent or incomplete, run LLM extraction on the HTML.
4. Auto-heal merge — If Schema.org is partial (e.g. missing price), merge with LLM-extracted fields to create a hybrid result.
5. Playwright fallback — For JS-heavy pages that return empty HTML, render with a headless browser first, then extract.

Cost per Method

Auto-Heal Merge

When Schema.org extraction returns partial data (e.g. title and price but no description), the router automatically:

1. Keeps all Schema.org fields at their high confidence baseline
2. Runs LLM extraction on the same HTML
3. Fills in missing fields from the LLM result at LLM confidence levels
4. Marks the result as hybrid with per-field provenance

This gives you the best of both worlds: high-confidence structured data where available, with LLM-powered gap filling.