Gemini OCR

The idea

The free tier limits how many requests you can make, not how big each one is. So Gemini OCR sends five pages per request instead of one, which cuts quota use fivefold, and keeps working through whole shelves of books.

Why it exists

To read large collections of Gujarati books without paid services, and without a local model that is limited by GPU memory.

What it does

• —Many requests at once, paced to stay inside each key’s rate limit
• —Spreads work across several keys from separate projects
• —Resumes where it stopped — safe to stop, including across days as the quota resets
• —Pages that are blocked or low-confidence get a second opinion from a different model
• —Spot-checks random pages by proofreading them, and flags differences without changing anything
• —Writes output in the format the knowledge base reads