# Pallyapp: WhatsApp AI accountability coach

An AI accountability coach that lives in WhatsApp. It learns your daily tasks, checks in on its own, and turns up the pressure when you go quiet.

*Daniel Pallant · Pallyapp · United Kingdom · 2026 - present · 2 months, then ongoing*

## The brief

Daniel wanted to turn accountability coaching into a product: a subscription service where users message a coach on WhatsApp, no app to install, set their tasks for the day, and get chased until they're done.

It had to feel like a persistent human coach rather than a chatbot, which meant starting conversations on its own schedule, remembering context across days, escalating when ignored and backing off when told. And it had to run unattended as a real business, with free trials, billing and admin oversight built in.

## What we built

- **WhatsApp coach.** onboards, plans and chases entirely in chat
- **Daily rhythm.** morning kickoff, mid-day nudges, evening wrap-up
- **9 AI tools.** “done with the gym” actually updates state
- **Escalation ladder.** pressure builds, never sounds canned
- **Natural-language pause.** “I'm busy till 3” just works
- **Stripe billing.** trial to subscription, handled end to end
- **Admin dashboard.** live users, subscriptions and usage

## How it works

A single Node.js service receives Twilio webhooks for inbound WhatsApp messages and Stripe webhooks for billing, with signature verification on both. Three timezone-aware scheduled jobs drive the proactive side, so everyone gets their 7am kickoff at their own 7am. Conversations run through a two-tier AI pipeline: a fast model classifies intent and extracts structure, then the full model writes the coach's reply with tool-calling against the task list.

## Stack

- **Bot:** Node.js, TypeScript, Express, MongoDB + Mongoose, Twilio WhatsApp API, OpenAI GPT models, Stripe
- **Scheduling:** node-cron, Luxon
- **Quality:** Zod, Winston, Vitest
- **Dashboard:** Next.js, React, Tailwind CSS, shadcn/ui
- **Infra:** Railway

## Numbers

- **24/7** autonomous accountability coaching
- **100%** message delivery
- **4,500+** messages exchanged in 4 months
- **1,000+** automated test cases across 84 suites
- **3:1** test-to-code ratio, on purpose
- **9** tools the coach can invoke on its own

## The hardest part

Making an LLM behave like a reliable employee instead of a chatbot. Anyone can wire GPT to WhatsApp. The hard part is everything around it: guaranteeing the morning message always fires even when pause logic, billing state and timezone math intersect in odd ways, stopping the model from confidently claiming a task was completed when it wasn't, and keeping escalating nudges human-sounding after the fifth one.

That is why the test suite is three times the size of the codebase: a large chunk of it replays real conversations end to end, so every one of those edge cases stays covered as the product grows.

## What the client said

> "I came to Roberto with nothing more than an idea and a vision. He took what was a jumble of thoughts in my head and turned it into a polished, fully functioning product. Nothing was ever too much trouble, and whenever I didn't understand something, he explained it in a way I could easily grasp. As a gas engineer rather than a tech expert, that made a huge difference. His technical knowledge, patience and strategic thinking have been outstanding, and I genuinely couldn't have asked for a better development partner. I've already trusted Roberto with further projects and will continue to work with him in the future."
> Daniel Pallant, United Kingdom, rated 5/5

More work: https://dock30.com/work

## Contact

- Email: crew@dock30.com
- Book a call: https://dock30.com/contact