SafeSkill

Inspiration

My uncle had Down's syndrome, he was the light of our family. Watching him navigate a world that wasn't built for him showed me how critical it is to have the right support at the right time. Therapy techniques like CBT can be transformative, but the way they're delivered often assumes a level of abstraction that excludes people with intellectual disabilities.

I wanted to build something that meets people like my uncle where they are, something warm, playful, and safe.

What it does

SafeSkill is a gamified therapy skill-building app for adults with intellectual disabilities. Clients practise evidence-based CBT skills - deep breathing, cognitive reframing, grounding ($5\text{-}4\text{-}3\text{-}2\text{-}1$ technique), and more - through interactive roleplay scenarios set across real-world contexts (home, work, the doctor's office, etc.).

The signature feature is the Safe Person Voice: a trusted voice (a celebrity or their own caregiver) narrates every step, turning a clinical exercise into something that feels familiar and secure. Therapists get a full dashboard to monitor progress, assign skills, and write session notes with AI-powered recommendations.

The app is structured as two interfaces sharing one codebase: a Super Mario-style world map for clients (5 worlds, 12 levels, star ratings, XP, streaks) and a clinical dashboard for therapists.

Challenges I faced

• Balancing clinical rigour with fun. Every skill module is grounded in CBT, but it had to feel like a game - not homework. Getting that tone right across $12$ scenario levels took many iterations.
• Accessibility at every layer. Our users have intellectual disabilities, so every interaction - button sizes, language complexity ($\leq$ Grade 3 reading level), audio narration - had to be carefully considered.
• Safe Person voice timing. Synchronising TTS narration with interactive UI steps so the voice never talks over user actions was trickier than expected.
• Therapist dashboard data design. Presenting per-skill accuracy, streaks, and engagement trends in a way that's immediately actionable for clinicians required close attention to information hierarchy.

What I learnt

• Inclusive design isn't a feature - it's a foundation. Building for cognitive accessibility from day one produced a better experience for everyone.
• Small rewards matter. The star system and streak counters drive engagement far more than we anticipated.
• The emotional weight of a familiar voice is enormous. Even in a prototype, hearing a "safe" voice walk you through a stressful scenario changes the entire feel of the interaction.

What's next

• Real caregiver voice cloning (with consent) to replace browser TTS
• Multilingual support beyond the current language selector
• Longitudinal outcome tracking aligned with standardised clinical measures (e.g., $\text{ABC-2}$, $\text{DBC-2}$)
• Pilot study with disability service providers

Built With

• elevenlabs
• next.js
• tailwind
• zustand