Overview

Therapair connects LGBTQI+, neurodivergent, and trauma-informed therapy seekers with practitioners who truly understand them. Prequel partnered with Unison Mental Health to design and ship an MVP: a conversational, privacy-first experience live as both a standalone product and a Unison-branded widget showing only Unison therapists.

Timeline: 6 weeks (Oct 2025) | Partner: Unison Mental Health | Stack: React · Cursor · Claude Code · ChatGPT · Figma Make · Notion API · GitHub · Hostinger

Challenge

Directories often prioritise logistics over people and rely on tracking that erodes trust. Seekers and therapists needed an experience grounded in empathy and privacy.

Key problems

• Limited filters for identity and lived experience

• Flat, form-based interactions that drive drop-off

• Third-party tracking undermining confidentiality

Goal

Design a therapist-matching flow that feels personal, affirming, and safe — scalable for partners and aligned with Therapair’s ethics of inclusion and privacy.

Approach (human-centred, AI-native)

Concept & Validation

Stakeholder interviews with practising therapists and Unison leadership validated the core problem, consent requirements, and tone of voice.

Discovery & Research (with co-design workshop)

We ran a focused co-design workshop (therapists + Unison) to align on user needs, language, and consent boundaries. Outputs included:

• End-to-end journey map (discovery → matching → contact/consent → handover)

• Consent touchpoints and plain-language microcopy principles

• Inclusion filters and accessibility priorities

• Risk/assumption log and a prioritised MVP backlog

These artefacts directly informed our strategy-as-prompts for ChatGPT and the component taxonomy in Figma Make.

Design & Prototyping

• Early concept test (Typebot): Checked comprehension and friction before high-fidelity work.

• Prompt-led design (ChatGPT → Figma Make): We wrote structured design prompts to guide Figma Make output — naming components, defining variants/states, accessibility tokens, microcopy tone (“supportive, plain English”), and mobile-first layout constraints.

• Prototype & iterate: Built interactive screens in Figma Make, then re-tested journeys (discovery → matching → consent/contact) and refined copy, spacing, and tap targets.

AI-Native Development

Dev environment (Claude Code in Cursor): Used Claude Code to scaffold a clean React workspace in Cursor (routing, env handling, a11y helpers, form state, Notion/SMTP utilities).

• Build the MVP:

• Unison therapist widget (embeddable, brand-themed)

• Landing page (clear value prop, trust cues, consent-first copy)

• Lead capture (minimal fields, one-question-per-screen)

• Smart email responses (templated confirmations + therapist handover; no tracking, clear privacy language)

• Notion database connections (secure insert/update, basic schema for therapists/enquiries, rate-limited writes)

Source control & deploy (GitHub ↔ Hostinger, coordinated in Cursor):

• Cursor orchestrated coding sessions, commit messages, and PR descriptions; Claude Code assisted with diffs and refactors.

• GitHub protected branches and PR reviews guarded quality; Hostinger pulled from the main branch or received CI artefacts for deployment.

• Secure API keys were injected at deploy via Hostinger environment variables (never committed), with .env in .gitignore, scoped tokens (least-privilege Notion integration, SMTP credentials), and a lightweight rotation note in the runbook.

• Documentation: Auth, consent copy, data-handling notes, CI/deploy steps, and an iteration runbook — generated and refined with Claude Code for clarity and consistency.

Design highlights

• Conversational prompts (“What brings you to therapy?”) over admin language

• One-question-per-screen to reduce cognitive load and drop-off

• Warm, minimal interface with accessible contrast and large tap targets

• Modular design system enabling quick partner theming (colour, type, components)

How we worked with AI (and why it mattered)

• Workshop → prompts: Co-design decisions became structured markdown prompts (goals, constraints, tone, a11y rules).

• Prompt → prototype: Figma Make generated coherent components/flows from prompts, accelerating high-quality exploration.

• Prototype → production: Claude Code translated patterns into stable React components and utilities inside Cursor.

• Coordination: Cursor acted as the “conductor” between local work, GitHub PRs, and Hostinger deployments — keeping changes auditable and fast.

• Impact: Shorter loops, more stakeholder time, clearer consent UX, and tighter partner fit — without bloating scope.

Privacy by design

• No Google Analytics or marketing pixels

• Self-hosted email (Hostinger SMTP)

• HTTPS and environment-variable encryption

• Partner data stored locally per organisation

• Secrets hygiene: tokens kept out of Git; least-privilege scopes; rotation guidance in docs

• Consent language front-loaded; no dark patterns

Build & delivery (at a glance)

• Repo: GitHub (protected main, feature branches, PR reviews)

• CI/CD: Simple build script + Hostinger deploy; status noted in PRs

• Secrets: Hostinger env vars; .env not versioned; per-env key sets

• Monitoring (lightweight): Health pings and email delivery checks (no tracking/pixels)

• Runbook: Release checklist, rollback steps, key rotation procedure

Impact

The MVP shows that AI-native methods and human-centred design can coexist to deliver respectful, emotionally intelligent experiences. It validates Prequel’s Workshop → Prompt → Prototype → Production model as a faster, higher-fidelity path from strategy to shipped product — without compromising empathy or ethics.

Next steps

• Broaden user research across Victoria (diverse cohorts)

• Refine matching logic and accessibility for wider rollout

• Add therapist dashboards and feedback tools

• Explore group and relationship therapy pathways

Credits

Concept, Design & Development: Prequel

Partner: Unison Mental Health

AI Tools: ChatGPT (design prompts), Claude Code (Cursor)

Infrastructure: GitHub (source), Hostinger (deploy)

Location: Melbourne, Australia