Search across a city council's public documents

Your city council publishes its agendas, meeting minutes and reports as PDF files on its website. The archive goes back to 1995, with 120,000 files and 2 million pages. About 40% of the pages are scans of paper, with no text in the file. The council adds about 300 files a week.

Residents and reporters want to search the archive for words, such as a street name. Each result shows the file's title and date, the page number and a few lines around the words. Clicking a result opens the file at that page. Searchers can also filter by committee and by year.

About 20,000 searches run a day, up to 5 a second. A new file should be searchable within a day of being published.

How does a PDF become searchable, and what does a search return?