I
Fast, model-free extraction
Parses dense pages in about 13ms with no GPU and no model warmup. A 300-page textbook becomes structured HTML in seconds.
❧13ms / page
A reading garden
Where dense PDFs blossom into reading.
ReadBlossom reopens cramped two-column papers, thousand-page textbooks, and complex reports as a calm, single-column page — warm paper, real typography, nothing to pinch or zoom. 13ms a page, no AI models in the loop.
13ms
per page
99%
text recall
0
cloud lock-in
Built for the PDFs that are hardest to read on a screen
arXiv papersTextbooksRFCs & specsGovernment reportsMagazinesTrade books
Core capabilities
Built for speed, fidelity & comfort
Traditional OCR and heavyweight models take seconds per page. ReadBlossom's deterministic parser reconstructs layout geometry instantly — and keeps it readable.
II
Semantic structure, kept
Heading hierarchy, multi-column flow, complex tables, footnotes, equations and captions are preserved, not flattened.
❧88.5% structure score
III
Reads anywhere
A responsive single-column view with dynamic type scaling, line-height tuning, and instant Paper / Sepia / Night themes.
❧3 reading themes
IV
Private by design
Runs in memory; the engine is model-free and can be self-hosted. No research or drafts sent to third-party AI APIs.
❧no cloud AI
V
A quality score, every time
Each conversion gets a 0–100 confidence rating from font consistency, reading flow and table alignment — so you know what to trust.
❧0–100 confidence
VI
Multilingual precision
ftfy normalization and wordfreq hyphenation repair handle German compounds, French accents and complex math symbols.
❧Unicode 15
How it works
From unreadable PDF to reading flow
Three quiet stages take your raw document and set it, like type on a press, into something made to be read.
I
Upload & triage
Drag in any PDF. In under a second a pre-flight pass checks encryption, scan status and structural integrity.
Pre-flight
status: digital · 42 pages · 2.4MB
II
Deconstruct the layout
At 13ms a page the engine categorizes blocks, rebuilds multi-column order, pairs captions and repairs hyphens.
Pipeline
pageindex → blocks → tables → refine
III
Read in full blossom
Open the reading view with adjustable type, inline footnotes, collapsible tables and saved progress.
Reader
theme: sepia · serif · progress 0%
For every reader
Made for anyone who reads seriously
Whether you're prepping for board exams, writing a literature review, or settling in with a classic novel.
For students
Turn 800-page textbooks and lecture slides into clean chapters you can highlight, search and read on your phone on the way to class.
- Continuous single-column scroll
- Export clean Markdown notes
- Dictionary & glossary links
For researchers
Digest arXiv papers, IEEE transactions and technical RFCs without fighting tiny two-column type or scattered references.
- Inline citation popovers
- Scrollable structured tables
- Formula & code fidelity
For book lovers
Breathe life into DRM-free archives, vintage scans and old manuscripts with elegant headings and warm, sepia pages.
- Cream & sepia palettes
- Custom serif & spacing
- Full-screen immersion
“A book is a garden
carried in the pocket.”
Arabic proverb
Open benchmarks
Tested on real documents, not demos
Every release is scored against a reference pipeline on an 11-document corpus — papers, books, manuals, reports and slide decks. Here are three of the hardest.
Technical book
A full programming manual with code listings, nested lists and hundreds of figures — converted end to end into a clean, scrollable view.
90.3
structure
99.8%
text recall
Pro Git — 501-page manual
from the public benchmark corpus
Trade book
A complete trade book: chapter structure, footnotes and every table reconstructed — reading order preserved at 98% across the volume.
92.6
structure
99.5%
text recall
The Psychology of Money
from the public benchmark corpus
Technical spec
A monospace specification with deep heading hierarchies and inline tables — heading levels detected at 96%, furniture stripped cleanly.
91.8
structure
99.6%
text recall
IETF-style RFC / spec
from the public benchmark corpus
13ms
Processing time
average per page
88.5%
Structure accuracy
across complex PDFs
11+
Document profiles
papers, books, slides
100%
Model-free
no neural nets, no GPU
Let your next PDF blossom.
Point ReadBlossom at the document you've been avoiding — the two-column paper, the scanned manual, the thousand-page reference. Read it tonight. No credit card required.
Start Reading FreeNo credit card requiredFree tier availableRuns in your browser