Summra Book

The problem

The classics are full of ideas worth reading. But Victorian sentence structure, archaic vocabulary and paragraphs that run for half a page keep most readers away. People who would love Wuthering Heights or Moby Dick never make it past chapter two.

Existing "modernized" editions either butcher the original or skip the hard parts. I wanted something different: the full original text, preserved, and right next to it a faithful modern translation, paragraph for paragraph, free for everyone.

Side-by-side reading

Every book on Summra has both versions. Read the original alone, the modern alone, or both side by side and jump between them. Paragraph boundaries match exactly, so you never lose your place when you switch.

The translation isn't a summary or a paraphrase. Every sentence in the original has a corresponding sentence in modern English, and the meaning, tone and structure are preserved.

The translation rules

"Modern English" is vague, so the pipeline holds every chapter to a concrete target: Flesch-Kincaid grade 7–9, the level most U.S. news sites aim for.

• Replace archaic vocabulary. Thee, wherefore, betwixt, countenance become you, why, between, face.
• Break up long sentences. A 60-word sentence with three semicolons becomes two or three shorter ones.
• Modernize idioms. Obscure references get their closest modern equivalent.
• Preserve meaning, tone and structure. Paragraph boundaries stay put; names and dialogue stay intact.
• No summarizing, no skipping. Nothing is dropped because it was "implied".
• Keep period detail. Characters still ride in carriages, not Ubers.

A validator checks all of this automatically: paragraph counts must match, the character ratio must stay between 50% and 110%, and the output must end on a sentence terminator. Anything that fails goes back through a self-healing pipeline.

By the numbers

• 90 classic books ingested from Project Gutenberg, 38 with full plain-English translations.
• 4,159 chapters parsed and aligned.
• About 16 million characters of plain-English prose, roughly 2.6 million words.

How it's built

• Backend: Python and Flask, with SQLite for storage.
• Translation: Google Gemini, with a multi-pass pipeline that enforces paragraph-by-paragraph alignment and re-runs chapters that come back truncated or summary-shaped.
• Quality control: a deterministic validator plus a self-healing orchestrator backed by Claude that diagnoses failures and dispatches the right fix.
• Frontend: a vanilla JS reader with Playwright end-to-end tests.

The hardest part

Not the translation itself, but getting paragraph boundaries to match exactly across 4,000+ chapters. LLMs love to merge or split paragraphs silently, and a single off-by-one breaks the side-by-side view. The pipeline has six steps just for paragraph alignment.