Building TotTrack: A Child Development App From Scratch

TotTrack started with a practical question: how can parents keep their observations, milestones and activities together between routine development reviews?

The answer was to build something purpose-made. A web and mobile app that lets parents log milestones, run activities, generate powerful development reports using AI, and share progress with family, nursery or medical practitioners — all in one place. Here's how we built it.

The Problem

Parents of children aged 1–5 may want to keep notes about milestones, play and questions to raise at a review. An organised record can support that conversation, but an app cannot determine whether a child has a developmental condition.

The existing options weren't great:

• Generic milestone checklists — static PDFs that don't adapt to your child
• Baby apps — focused on feeding, sleeping and nappy tracking, not development
• Scattered notes — photos, questions and observations spread across different places

We wanted to combine a parent-held record with ideas for activities. The NHS explains how health and development reviews work and advises parents to contact their health visitor or GP if concerned; they do not need to wait for the next review. TotTrack is not a diagnostic or screening tool and does not replace those services.

What We Built

Milestone Tracking

Parents log milestones as they happen or retrospectively, with photos and notes. These records help families remember observations and prepare questions for a professional; milestone timing alone is not a diagnosis.

54 Developmental Activities

The app includes 54 structured activities across 7 developmental categories: cognitive, creative, language, motor, sensory, social, and custom activities parents can create themselves. Each activity is age-appropriate and designed to be done at home with everyday items.

AI-Powered Development Reports

Using the milestone and activity data a parent has logged, TotTrack creates AI-generated summaries powered by Claude. These reflect the information entered by the parent and may contain errors or omissions; they are not clinical assessments.

Reports organise the recorded observations by developmental area and can be downloaded as PDFs to discuss with a nursery or health visitor. Families should check the summary against their original notes before sharing.

Personal Development Plans

After each analysis report, the system automatically generates a development plan. This recommends specific in-app activities the parent can do, plus external activities and resources. The plan is tailored to the child's current stage and the areas flagged in their report.

Family Sharing

Parents can invite partners, grandparents, or carers to view their child's profile. Role-based access means you can control who sees what — full editing access for co-parents, read-only for grandparents who just want to see the latest photos and milestones.

The Tech Stack

We chose a modern stack that prioritised speed of development and cross-platform reach:

• React + Vite + TypeScript for the web app — fast build times, type safety, excellent developer experience
• React Native + Expo for iOS — shared component logic with the web app
• Tailwind CSS for styling — rapid UI development without fighting custom CSS
• Drizzle ORM + Neon PostgreSQL for the database — serverless Postgres with type-safe queries
• Anthropic's Claude for AI reports and development plans — used to summarise parent-entered observations
• Stripe + Apple IAP for billing — 14-day free trial, then subscription

Hard Problems We Solved

AI Report Reliability

When you're generating multi-page development reports with AI, reliability is everything. Parents are reading these reports to understand their child's development — they need to be accurate and well-structured every time.

We built a pipeline with structured prompts, output limits, JSON sanitisation and a repair step for malformed output. These checks improve formatting and delivery, but do not establish medical accuracy or clinical validation.

Cross-Platform Without the Pain

Running a React web app and a React Native iOS app from the same codebase sounds great in theory. In practice, it means carefully managing which components are shared and which need platform-specific implementations. We ended up with a pragmatic split: shared business logic and API layer, with platform-specific UI where the experience needed to differ.

Background Processing

AI report generation is an intense process and takes a few minutes. You can't make a parent stare at a spinner for that long. We built a background job system that kicks off report generation asynchronously, tracks progress, and notifies the parent when their report is ready. If the server restarts mid-generation, stale jobs are automatically detected and marked as failed so parents aren't left waiting forever.

What We Learned

A few takeaways from building TotTrack that apply to any custom app project:

• Start with the core loop. For TotTrack, that's: log a milestone → run an activity → generate a report. Everything else is secondary. We shipped with this loop working end-to-end before adding family sharing, billing, or development plans.
• AI features need guardrails. LLMs are powerful but unpredictable. Every AI-generated output needs validation, error handling, and a fallback. We invested heavily in this and it paid off.
• Cut scope ruthlessly. We originally included emotion tracking, mood logging, nutrition tracking, and sleep tracking. All removed. They diluted the core value proposition and added maintenance overhead without proportional user value.
• Error monitoring from day one. We built a self-hosted error logging system that captures server errors, client errors, and background job failures. A cron job emails summaries every two hours. This caught issues in production that would have taken days to surface otherwise.

The Result

TotTrack is available at tottrack.co. The build demonstrates how a focused product can bring records, sharing and AI-generated summaries into one workflow. We have not published a clinical evaluation or independently verified user-outcome study.

More importantly, it shows the kind of application we build at SH Applications. Not templates. Not no-code prototypes. Real, production-grade software designed around a specific problem, built to last.

Reviewed and updated 1 October 2026.