Site Scope

The problem

Researching graduate programmes means reading dozens of admissions sites that all hide the same twelve facts in different places. I wanted the facts, not the browsing.

How it works

• Renders pages in Playwright Chromium and simulates scrolling to surface lazy-loaded content, falling back to plain HTTP when a browser isn't available.
• Strips navigation, headers, footers, sidebars, forms and scripts before analysing text, so the extractor reads content, not chrome.
• Classifies links into application portals, official documents, faculty contacts, financial aid and internal subpages.
• Quick Scan reads the current page; Deep Scan follows up to five relevant subpages.

A deliberate choice

Extraction is deterministic and source-grounded rather than an LLM summary. When a deadline matters, you want to know exactly where it came from. The normalised output is shaped so a model can sit on top later without touching the scraping layer.