Case study · BlurbBio

A writing studio that reads the whole manuscript.

BlurbBio is an AI platform for fiction authors. It keeps a novel's characters, canon and timeline in a Story Bible, checks each chapter against it, and takes the finished book through to print-ready files.

Visit blurbbio.com
Challenge
Novels are too long for chat tools, and authors juggle separate apps to write, check and publish.
What we built
One studio with a Story Bible, whole-book AI analysis and print-ready export.
Result
120+ writers now manage 1.4M+ words in BlurbBio.
app.blurbbio.com
Type
AI writing platform
Status
Live in production
In development since
August 2025
Platform
Web app
Stack
Next.js, Supabase, Python

120+

writers using BlurbBio

84

book projects

1.4M+

words managed

10

manuscript formats imported, including Scrivener projects

50 of 51

canon findings correct in a manual spot-check

1.8 s

from clicking a book to editor text on a clean live test

Usage from BlurbBio's production dashboard, excluding test accounts. Accuracy and speed measured on test manuscripts, in October 2026.

01

The problem

A novel is too long to hold in your head and too long for a chat window. A 70,000-word draft carries hundreds of small commitments: who knows what, who is dead by which chapter, what colour someone's eyes were in chapter one. Readers notice when those slip.

General AI assistants see a fragment of the book at a time, so their feedback is confident and hard to check. Authors also end up moving the same manuscript between a writing tool, continuity notes, a formatter and a cover tool before anything reaches a printer.

The goal was one place to write, check and publish a book, where anything the AI said could be traced to the author's own text.

02

Who it's for

BlurbBio is built for people finishing long fiction, not for people generating it.

Novelists in revision

Writers with a full draft who need continuity, pacing and structure checked across the whole book.

Self-publishing authors

Authors who need interior files, covers and ebooks that meet the specifications of print platforms such as Amazon KDP and IngramSpark.

Authors with existing manuscripts

Writers bringing work in from Word, Google Docs or Scrivener, who need structure and formatting to survive the move.

03

What we built

How a chapter-aware check reads a bookInterface illustration
Chapter 1Chapter 50
Ch 12 · FlashbackSet before her death. Not an error.
Ch 43 · Hannah diesRecorded as a change at this chapter.
Ch 47 · Alive againFlagged, with the sentence quoted.

Manuscript editor

A focused editor with chapters, scenes and outline side by side, autosave, and named versions an author can return to.

Story Bible

Characters, places, relationships, timeline, canon rules, arcs and foreshadowing, kept in step with the text. Sync proposes changes; the author approves them in a review inbox.

Live detection

A Health view that flags continuity conflicts, canon breaks and timeline errors, scoped to the chapter and scene where they occur.

Seven AI analysis tools

Story Doctor, Behavioural Consistency, Dialogue, Scene Purpose, Subplot, Drop-off and Genre, each reporting on the whole book.

Copilot and Brainstorm

Writing assistance that reads the Story Bible and knows what state each character is in at the current chapter.

Import Studio

Ten formats, a structure preview before commit, split, merge and rename controls, and one-click undo.

Publishing

Print interiors, full cover wraps sized from the real page count, EPUB validation and a single export package for a print platform.

Launch Kit

Launch materials and a cover designer, kept alongside the book they belong to.

Intelligence panel showing a manuscript health score, submission readiness and scores for plot, character, dialogue, pacing, structure and genre
The Intelligence panel on a test novel, combining the AI tools into one view of a manuscript.
BlurbBio's import review showing 69,831 words split into 50 chapters and 107 scenes before anything is saved
Import review on a test novel. Authors check the detected structure before anything is saved.
Story Bible web view showing characters, places and objects from a novel connected by their relationships
The Story Bible as a web. Every thread is a stored relationship, none is guessed.
Story Doctor result separating AI judgment from measured counts, with a bar showing how many findings quote the manuscript
Story Doctor keeps AI judgment and measured counts apart, and shows how many findings quote the text.

04

Architecture

The web app handles everything an author touches. Work that runs for minutes, such as reading a whole novel, parsing a manuscript or rendering a print file, moves to Python services that the app starts and polls as jobs.

Client

  • Next.js and ReactDashboard, tools and publishing flows
  • Tiptap editorManuscript editing, with save gating

Application

  • API routes on VercelOwnership checks on every project route
  • Plan and credit gatesReserve first, charge on delivery
  • Scheduled sweepsBackground Story Bible sync, job recovery

Python services

  • AI analysisThe seven whole-book tools
  • Bible scannerPer-chapter extraction, book synthesis
  • ImportParser with memory and rate limits
  • PrintPDF rendering, EPUB validation
  • NLPMeasured text statistics

Data and models

  • SupabasePostgres, auth and file storage
  • Upstash RedisFast shared state
  • OpenAI by roleBulk, scan, verify and judge tiers

05

Key engineering decisions

1

Run whole-book work outside the web tier

Why

Reading a novel can take many minutes, far past a serverless request. Job-based Python services let those runs finish and report progress.

Trade-off

Two deploy surfaces. Some releases have to go out in a set order, because a newer print service rejects requests from an older app.

2

Findings must quote the manuscript

Why

A verifier checks each AI finding against the text and marks what it could not ground. Measured counts and AI judgment are labelled separately in the interface.

Trade-off

Fewer findings reach the author, and each run costs more than simply trusting the model.

3

Authors approve what enters their Bible

Why

Only evidence and character presence are accepted automatically. Everything else waits in a review inbox, so the AI can't quietly rewrite a book's facts.

Trade-off

Authors have an inbox to work through. We treated that as the correct cost of accuracy.

4

Ship dark until a feature clears a measured bar

Why

We keep gold-standard answer keys for test manuscripts, including a control book with deliberately planted errors. A check goes live only when it passes.

Trade-off

Some finished features stay off. Every risky change also carries a kill switch, which is more code to maintain.

5

Choose models by role, not by product

Why

A cheaper model does the bulk reading, a stronger one judges candidates. One configuration per service makes it possible to swap models and re-run every evaluation.

Trade-off

More moving parts per analysis, and each model change means re-measuring.

06

The hardest problems

Telling a continuity error from a story

Problem

A Story Bible with one status per character can't express "alive until chapter 43". Checkers flagged flashbacks and earlier scenes as mistakes, and authors stop trusting a tool that cries wolf.

What we did

We made the Bible chapter-aware. Each character's state is resolved per chapter from recorded changes, flashbacks are scoped by when a scene is set, and Sync proposes a change at a chapter instead of overwriting the base record.

Result

In a mutation test on a full-length novel, all three planted errors were caught with no false flags.

07

What shipped

Live today

  • Editor with scenes, autosave and named versions
  • Story Bible with sync, review inbox and chapter-aware state
  • Live continuity and canon checks
  • Seven whole-book AI analysis tools
  • Copilot and Brainstorm
  • Import Studio with ten formats, including Scrivener
  • Print and ebook export, cover designer and Launch Kit

Still to come

  • Checks that haven't cleared their bar, such as a subplot sweep for dropped threads, stay switched off until they do
Next case studyAskOrac

Let's build.

Tell us about your project. We respond within 24 hours with a detailed plan of attack.

Start a Project