Saraan Asim
Saraan Asim
Stage
0→1
Product
AI publishing platform for authors
Dates
Oct to Dec 2025
My part
Built the architecture, backend and AI features
Stack
React, Express, Drizzle, OpenAI, ElevenLabs, Firebase, Stripe
Live
chattybook.com

ChattyBook

An author-first publishing platform that turns manuscripts into interactive reading, with voice narration, AI conversations about the book and support for several languages.

ChattyBook's home page with the headline "The Author-First Publishing and Marketing Platform" and a looping video beside it.
Fig. 1. ChattyBook's public home page.

0→1

The product

ChattyBook is an AI publishing platform built around the author. It turns a manuscript into interactive reading: readers listen to voice narration, ask the book questions in an AI conversation, and read in more than one language. Authors publish the same book as an e-book, an audiobook and an AI chat, earn directly and keep their rights.

What I built

Architecture and backend. I built the architecture and the backend on Express with Drizzle, with Firebase and Stripe.

The manuscript pipeline. The pipeline that takes an author’s manuscript and turns it into a book for readers, split into chunks with embeddings so the rest of the product can work across the whole text.

Voice narration. Narration with ElevenLabs, so every book can also be listened to.

AI conversations and translation. Readers ask the book questions and get answers grounded in its text, and the reading experience supports several languages.

How it works

Readers use a React app and authors upload manuscripts through their own tools. Both call one Express API that stores books, chunks and conversations through Drizzle. A question from a reader goes to the API, which finds the most similar chunks of the book, adds the spoiler-free context for the reader’s current page and the recent conversation, and streams the answer back from OpenAI. ElevenLabs produces the narration, Firebase supports the app, and Stripe takes payments.

3 decisions

Decisions

  1. Built for long manuscripts

    The hardest constraint was length. A whole book can't go into one AI request, so manuscripts are split into chunks with embeddings. When a reader asks a question, the closest chunks are found by similarity and passed to the model with the recent conversation.

  2. Answers that don't spoil the book

    Along with the text chunks, the AI gets character profiles, concept profiles and chapter summaries for the book. They are filtered to the page the reader has reached, so a question about a character never gives away what happens later.

  3. A conversation style per kind of book

    Each book uses an AI style, educational, fiction, mystery or custom, and the system prompt is built from it along with the reader's name, the title and the author.

Feature level

From the project record

Architecture

Reader appReactAuthor toolsmanuscripts inAPIExpress · DrizzleOpenAIconversationsElevenLabsnarrationFirebaseStripepaymentsReader appReactAuthor toolsmanuscripts inAPIExpress · DrizzleOpenAIconversationsElevenLabsnarrationFirebaseStripepayments
Fig. 2. One API turns a manuscript into a book readers can listen to and talk with.

Result

Outcome

ChattyBook is live at chattybook.com. Authors publish a book as an e-book, an audiobook and an AI chat in one place, and ChattyBook's offer to them is 70% of every sale and their readers' emails.

Looking back

What I’d change

I would cache the generated audio and text earlier. Narration and AI-generated text take time and money every time they are produced. Storing each result the first time and serving it from the cache after that would have kept both down from the start.