18: Document Rescue and OCR - transcript
Jessica: The QUILL Cast, episode eighteen. I'm Jessica, and today is the document rescue episode, the one for every PDF that turned out to be photographs of paper.
Liam: I'm Liam. Every listener knows this moment. The lease from two thousand eleven. The photocopied class handout. The court document from a scanner older than you. Your screen reader opens it and finds... nothing. Because there is no text. Just images of text.
Jessica: QUILL's answer is the Import slash Convert Document tool, top of File, Import, also under Tools, Reading and Dictation. It has one rule, and QUILL states it in exactly these words: free first, local first, and nothing is ever uploaded.
Liam: The pipeline has two free tiers. Pick any file, Word, PowerPoint, Excel, HTML, EPUB, PDF, or a straight image, a photo of a page. Tier one runs first: the built-in local converter, MarkItDown, which extracts the text that born-digital files already contain. For most files that's the whole trip: two seconds, clean editable text, done, and QUILL announces, nothing was uploaded.
Jessica: But the scanned PDF comes back nearly empty, and here's the design moment we love: QUILL measures what it recovered, and refuses to silently hand you a blank document. Instead it tells you the truth: QUILL could not find readable text in this document. It looks scanned or image-based. Run free on-device OCR? This stays on your computer and does not upload anything.
Liam: Three buttons: yes, run the local OCR. No, open the empty result anyway, sometimes you want confirmation of what's there. Or cancel. You decide, always, even though the OCR is free and private, QUILL still asks. That's the consent culture applied uniformly.
Jessica: Say yes, and tier two runs: real optical character recognition with the Tesseract engine, entirely on your processor. No internet, no graphics card, works on a plane. QUILL takes the PDF apart page by page, narrating progress, recognizing page three of twelve, and reassembles the text with page markers you can search, so page nine of the scan stays findable as page nine.
Liam: And then the honesty feature that separates this from every OCR tool we've used: confidence reporting. Tesseract knows how sure it was about each word, and QUILL tells you the average. A clean scan comes back with high confidence, trust it. A crooked fax from nineteen ninety-eight comes back flagged: confidence is low, review this result carefully. No false certainty, ever. You know whether you got a rescue or a rough draft.
Jessica: Setup is a one-time thing: the engine is a free download, about forty-eight megabytes, Tools, Reading and Dictation, Install Local OCR Engine. QUILL fetches the official installer from its own verified release storage, checks it byte for byte against a pinned fingerprint, and then opens the installer visibly for you to finish. QUILL never silently installs software. And if Tesseract already lives on your machine, from anywhere, QUILL just finds it.
Liam: There's also a plain-language services page, OCR and Conversion Services, that describes each tier like a human: what it does, what it's best at, what it costs, nothing, and whether the engine is installed. It also says, honestly, that a cloud tier for the hardest documents, dense tables, handwriting, is planned, strictly opt-in, with cost and upload spelled out first, and the free tiers always running before it.
Jessica: The workflows write themselves. Students: photograph the handout, feed the image in, read it on the bus. Anyone with paperwork: taxes, leases, medical letters, exactly the documents you don't want in a cloud, exactly the ones this keeps at home. And pair it with a watch folder, episode twenty-one, and your scanner's output folder becomes a self-reading pipeline.
Liam: Homework: find the worst PDF you own, the one you gave up on years ago, and run it through Import slash Convert Document. Let the escalation prompt do its thing. Check the confidence report. Then save the text that was locked away all this time.
Jessica: Next episode: files that live elsewhere, servers, GitHub, SSH, and the end of the download-edit-reupload shuffle.
Liam: I'm Liam.
Jessica: I'm Jessica. Free the words.