Case study · AI chatbot

Prologue Partnerships: an expert's book, turned into a paid AI coach

Prologue turns an expert's books, seminars and know-how into an AI companion that people pay to talk to. The first one is Oh Crap! Chat, built with author Jamie Glowacki on her potty-training method. A parent who is stuck at bedtime asks a question and gets Jamie's approach for their child, without rereading the book.

The Prologue Partnerships home page: Your expertise, always available, beside a sample chat with an expert guide
Visit prologuepartnerships.com
Client
Prologue Partnerships
First product
Oh Crap! Chat, with author Jamie Glowacki
Who it serves
Parents of children aged 1 to 4
Business model
Paid access passes, no subscription
Build time
About a year to launch
Core stack
React, Google Cloud, Gemini, Pinecone, Supabase, Stripe

For everyone

The short version

Parents in the middle of potty training have questions at odd hours. The answers are in Jamie Glowacki's book, Oh Crap! Potty Training, but nobody wants to flip through a book at 9 p.m. with a toddler who just had an accident on the rug.

So we put the book behind a chat window. Parents ask in their own words, and the bot answers with Jamie's method for their child's age and situation. When a question needs a specialist, like poop withholding, it says so and points them to the right help.

Parents buy a pass for 24 hours, 3 days or 10 days. There is no subscription to cancel. More than 700 parents have paid so far, and over 400 are active users.

  1. The author's method

    Jamie's book and teaching, loaded into a private knowledge base.

  2. An AI coach

    Answers from that material only, in plain words, at any hour.

  3. A parent with a question

    Pays for a short pass, gets help, and the author earns from it.

For founders and CEOs

The business case

What the business needed, why it was built this way, and what it has returned so far.

The problem

The answer exists, but it's buried

Parents know the book covers their problem. Finding the part that fits a 26-month-old who refuses the potty after a good first week is the hard part.

Help is needed outside office hours

Potty-training trouble happens at home, at bedtime and on weekends, when no consultant is on call.

One expert can't coach everyone

Jamie can work with a few families at a time. Many more readers need the same guidance, fitted to their child.

Why we built it

Prologue's idea is that an author's knowledge is worth more when readers can ask it questions. Each partner author gets a chatbot that learns only from their own material, and the author's audience pays to use it.

We started with one author and one subject on purpose. Potty training has a clear age range (1 to 4, with a sweet spot of 20 to 30 months), parents with urgent questions, and an author whose method people already trust. If it worked there, the same platform could serve the other fields Prologue lists: relationships, child development, pet training and coaching.

The admin side was built for that next step from day one. It already manages partner authors, their teams, bot presets, customers and revenue, so adding the second author is setup work, and the platform itself stays the same.

Prologue use cases: family and relationships, parenting and development, animal behaviour

How it makes money

Parents pay once for a block of time. Potty training is an intense few days and then the odd check-in, so a short pass fits how people actually use it, and nobody has to remember to cancel anything.

Authors bring their audience. Prologue brings the product, the paywall and the marketing.

Price per day of access

Prices from the Oh Crap! Chat checkout page. The 3-day pass is marked best value.

What it has returned

It took about a year of building to launch. Since then, more than 700 parents have paid for access, and over 400 of them are active users.

For Prologue, that settles the main question: parents will pay for an author's method in chat form. The next step is more authors, each with their own bot, on the same platform.

For engineers and technical leads

How it's built

How the system answers from one author's material only, sells time-limited access, and ships changes without breaking things for parents who have paid.

How Oh Crap! Chat answers a parentParentphone or laptopfilters bad trafficCloudflareDNS · CDN · WAFaskHTTPSGoogle CloudAPI on Cloud Runchecks the pass,builds the promptGemini on Vertex AI2.5 Pro · EU regionquestion + 8 passagesanswer, streamed backtop 8 passagesPineconethe author's private indexpass still valid?Supabaseaccounts · passes · chatsbuys a passStripe Checkout$10 · $15 · $40 passeswebhook: pass + end timeAdmin dashboardUpload. Chunk. Embed. Check.1 file kept in Cloud Storage2 split into chunks, turned into vectors3 written to Pinecone, every chunk counted4 live progress on screen (server-sent events)vectorsAuthor's book,notes and seminarsAnswer from the book,or say so.Withholding questions go toa specialist, never a guess.
How a question becomes an answer

  1. The parent opens the chat in the browser. Cloudflare sits in front of everything and handles DNS, caching and filtering bad traffic.

  2. Sign-in runs on Supabase, with Google, Apple or email. Payment runs on Stripe Checkout, and a webhook records which pass the parent bought and when it ends.

  3. Each question goes to the API on Google Cloud, which checks the parent's pass first. Without a valid pass, the model is never called.

  4. The API searches Pinecone for the passages from Jamie's material that best match the question, eight by default.

  5. Gemini on Vertex AI writes the answer from those passages, the bot's instructions and the recent chat, and streams it back as it writes.

  6. New material goes in through the admin dashboard. The file is kept in Google Cloud Storage, split into chunks, embedded and written to Pinecone, with live progress on screen and a check that every chunk arrived.

The stack

Front end

  • React (Vite build)
  • wouter routing
  • Cloudflare DNS, CDN and WAF

Back end

  • API on Google Cloud
  • Streamed chat responses
  • Server-sent events for upload progress

AI

  • Gemini on Vertex AI, 2.5 Pro by default (EU)
  • Pinecone vector search
  • Google Cloud Storage for source files

Accounts and payments

  • Supabase Auth: Google, Apple, email
  • Stripe Checkout and webhooks
  • Two-factor sign-in for admins

Operations

  • AI usage and cost dashboard
  • Error events and chat feedback (CSAT)
  • Cookie consent and analytics

Problems we had to solve

Answers that stay inside the book

A general chatbot will make up potty-training advice that sounds confident. Here the model only sees passages pulled from Jamie's material, and each bot has its own instructions about scope. Topics the method leaves to specialists, such as poop and pee withholding, are named on the checkout page, and the bot points parents to specialist help for them.

Selling time instead of a subscription

A pass that runs out after 24 hours, 3 days or 10 days means access has to be checked again on every message. Stripe's webhook writes the pass and its end time, and the API checks it before any model call, so expired passes don't use the model.

Streaming without leaving parents waiting

Answers stream to the browser as Gemini writes them. The API has one time limit for the first words and another for the whole answer, and it retries with a growing delay when the model is busy, so a slow moment at Google turns into a short pause on screen.

Tuning each bot without a release

Every bot has its own settings in the admin dashboard: Gemini model, region, temperature, how many passages to retrieve, how much chat history to keep, time limits and retries. Prologue can tune a bot for a new author without a code change.

Each author's material stays theirs

Prologue promises authors that their content is never shared with anyone else. Each bot reads only its own author's material, and partner authors and their team members have their own logins for their bots, customers and revenue.

Knowing what each answer costs

Gemini is billed by usage, so the dashboard shows AI usage and an estimated daily cost next to revenue and transactions. A $10 pass only works as a business when you know what a conversation costs to answer.

Shipping changes safely

From a code change to parents' screensGitHubpush to mainCloud Buildinstall · lint · test · buildStagingStripe in test modere-ask real parent questions,compare the answersLooks right?sign-offProductionCloud Run behind Cloudflarerollback to the previous revisionUntested builds neverreach paying parents.
From a code change to parents' screens

  1. Every change is built and tested on Google Cloud before it goes anywhere.

  2. It runs first in a staging copy with Stripe in test mode, so payments and passes can be tried end to end.

  3. A set of real parent questions is asked again to compare answers before release.

  4. Only then is it promoted to production. The previous version stays ready to roll back to.

Keep reading

Related work and services

Next case study