PDF Retype is an AI-powered document conversion platform that transforms PDFs into fully editable Word, Excel, and PowerPoint files while preserving formatting, tables, and layout.

The Problem

Manually retyping or reformatting PDF content is tedious, error-prone, and wastes hours of productive time. Existing tools either break formatting, miss tables, or require expensive subscriptions.

Our Solution

PDF Retype uses advanced OCR and AI to accurately extract and reconstruct document structure - including complex tables, multi-column layouts, and embedded images - into native Office formats.

Key Features

  • Convert PDF → Word (.docx) with full formatting preservation
  • Convert PDF → Excel (.xlsx) with table detection
  • Convert PDF → PowerPoint (.pptx) with slide reconstruction
  • Batch conversion support
  • Fast processing (most documents under 30 seconds)
  • Privacy-first: files deleted after processing

Built With

Python, FastAPI, AWS, Tesseract OCR, OpenAI API, React

Built With

Share this project:

Updates

posted an update —

Architecture update on PDF Retype: Preserving official notary seals, stamps, and signatures

We have deployed a major architecture enhancement to our TrueLayout reconstruction engine focusing on dual-layer document decoupling for legal deeds, certificates, and notarized agreements.

Conventional OCR tools either flatten official documents into locked, uneditable background images, or treat circular rubber stamps as text characters, injecting garbled alphanumeric noise directly into the middle of legal clauses.

Our dual-layer pipeline introduces:

  1. Chromatic deconvolution to isolate notary seals and ballpoint pen signatures from carbon toner.
  2. High-DPI transparent alpha-channel extraction to eliminate opaque white bounding box traps.
  3. Relative paragraph z-index anchoring, allowing 100% editable Word paragraph reflow without occluding authentication marks.

You can test live document conversions with intact seals and stamps at https://pdfretype.com.

posted an update —

Architecture update on PDF Retype: Solving borderless financial statement reconstruction

Over the past few weeks, we've tackled one of the hardest layout engineering challenges in document reconstruction: converting multi-page, borderless financial balance sheets and audit schedules from PDF into native Microsoft Excel and Word.

When naive converters process borderless tables, they treat the page as a 2D coordinate canvas. This leads to three widespread failure modes:

  1. Multi-line line items split into orphan rows, causing blank columns and formula misalignment.
  2. Silent non-breaking spaces (CHAR 160) between currency signs and digits cause Excel =SUM() formulas to return zero without warning.
  3. Negative values formatted in accounting parentheses (1,450.00) turn into uncalculable strings.

Our TrueLayout engine focuses on establishing vertical whitespace projection envelopes and unifying multi-line text blocks into single cell paragraphs before table generation. This ensures that extracted spreadsheets feature genuine, dynamic cell grids that expand naturally with edits and calculate out-of-the-box.

You can test the latest table extraction capabilities on live documents at https://pdfretype.com.

posted an update —

Quick progress update on PDF Retype:

Over the last few weeks we've been putting a lot of focus into fixing the most annoying part of PDF to Word conversion, which is broken table grids and random floating textboxes that jump around whenever you press enter.

Most traditional tools just slap coordinate boxes everywhere, which look okay visually at first glance but are a nightmare to edit . We have been tuning our layout reconstruction so tables actually convert into clean native Word tables with real cells and borders, and paragraphs stay in a continuous reading order instead of five separate disconnected blocks.

If you deal with scanned reports or messy document conversions, feel free to try it out at https://pdfretype.com and let me know if you run into any formatting quirks or tables that behave weirdly? Always open to feedback!