Daniel Cohen
liveJan '26 to now

Sparktale

Illustrated, personalised children's books, printed and shipped.

Why I built itI wanted children to learn by seeing themselves as the hero of a story, from a two-minute form instead of weeks of custom work.
  • AI can now make that book well enough to print.
  • Live since Jan '26, built with a partner developer.
Status
live
When
Jan '26 to now
Type
B2C web, print
My role
Two founders: I own product, marketing and legal coordination

The product question

Can we reach parents who will pay for a personalised book, and what does one sale leave after every cost?

Scope

Jan '26 to now
from the first commit in the code repo to today
5 languages
English, German, Hebrew, Spanish and Portuguese, all live in the app
sparktale.ai
Crop of the live Sparktale landing page: headline Spark a Tale They'll Never Forget, a short description, a Spark a Tale button and four short benefit labels

Screenshot of the live site, taken Oct '26.

What I tried

Custom children's books take weeks to make, so Sparktale turns a child's name, a photo and a two-minute form into an illustrated storybook with the child as the main character.

  • Books come as a printed hardcover or softcover through print-on-demand, plus a PDF, in five languages: English, German, Hebrew, Spanish and Portuguese.
  • A 5-step wizard collects the child, the book type, the theme, the art style and a custom touch.
  • Sign-up comes after the reader has seen the book.
  • The character stays consistent across pages and single illustrations can be regenerated.
  • A page-flip reader supports right-to-left for Hebrew.
  • I own product, marketing and legal coordination. A partner developer leads engineering.
  • I wrote a kill test before spending.
  • Prose lives in a knowledge-base repo, code-coupled docs in the code repo and binaries in Drive. Decisions and reviews are dated documents.
  • The compliance review was multi-jurisdiction. It was built from repo audits and jurisdiction research and red-teamed by separate agents.

Decision timeline

  1. Started with a second founder. I own product, marketing and legal coordination. A partner developer leads engineering.

  2. Wrote a kill test.

  3. Built the multi-jurisdiction review from repo audits and jurisdiction research. Separate agents red-teamed it.

  4. Checked live at sparktale.ai. It has paying customers.

Questions I asked along the way

  • Should the book jacket preview be on by default?

    Option rejected
    Default off, which would ship the old preview of painted sample pages and make the jacket opt-in. Any environment that forgot the setting would then run the old path without anyone noticing.
    What happened
    The brief recommended on by default, with a server-side switch that brings back three painted sample pages without a code change. The switch lets me restore the old preview; the brief records no conversion result.

What happened

Still running. Sparktale is live at sparktale.ai and has paying customers.

  • Reaching parents was the hard part.

What I took from it

  • A kill test written before the first euro of spend. It went into step 4, Make a decision, of the nine steps.
  • Dated decision records and a two-repo split for prose and code. Both are in the nine steps and their templates.
  • Work handed to a partner developer as issues that say what and why. That is the distill-to-gh-issue skill.
  • Growth depends on reaching parents, and one sale has to leave something after every cost: principle.