Tandem: Voice Practice for Children Between Therapy Sessions
Children practise the speech goals their therapist has set, at home between sessions, with a gentle voice character, a parent beside them and no audio kept.
A voice practice app for children between speech therapy sessions. A friendly character called Pip runs short sessions on targets the clinician has approved, such as asking for things. The child can answer by speaking or by tapping a picture. The clinician portal shows progress, flagged trials and incidents, and every programme change is a new version the clinician approves. No audio is stored.
Children with speech and language goals often see a therapist once a week. Practice between sessions depends on busy parents and is hard to keep consistent. Any AI in this setting has to be safe with a young child, defer to the clinician and keep very little data.
The practice engine follows a fixed script of allowed steps, with 11 safety rules: a parent present, a stop phrase, a halt on distress words or repeated silence, a cap of two corrections, session time limits, and no points, streaks or badges. The voice speaks only lines the engine allows. Agents can propose programme changes, and only the clinician can approve them.
The system runs real voice sessions for two synthetic learners, with picture-board input for children who do not speak. All 11 safety rules and every step of the script are covered by 42 automated tests. The clinician portal handles programmes, consent, incidents, an audit export and blinded rating of the coach notes.
- Status
- Working system
- Client
- Confidential
Built to the Governed Cognitive Layer: the model reads and drafts, deterministic checks decide what it may do, and a named person approves anything that commits the business.
When the user is vulnerable, write the safety rules first and test them as code before building anything clever. Let the AI suggest changes, and have a qualified person approve them. Keep only the data you need. Those three habits make a sensitive AI service defensible.
Concept paper and live demo
The concept paper and the working practice app sit behind a password. I walk clinicians and funders through it live.
●Live demo (on request) →The same design rules behind this build, applied to your business in half a day.
Book the WorkshopDiscuss a build like thisScreens from the system
Demo data throughout. Customer names, accounts and payees are fictional.
Other projects
Agentic AIOneness: A Private AI Companion with Safety Built Around It
Adults talk privately about work, relationships and decisions, by text or voice, with every message checked for risk before any reply and a human listener one consent away.
See how it works →
Agentic AIAgentic Tour Operations Platform for a Destination Management Company
Offshore staff and bounded agents prepare costed tour proposals from a WhatsApp enquiry, and the founder decides at three gates: release, final price and job confirmation.
See how it works →
Agentic AIGold Desk: Agentic Analysis for a Gold Trader
Four analyst agents, a bull and bear debate and a deterministic risk gate turn live gold market data into trade ideas a trader can check, with nothing sent to a broker.
See how it works →
The hardest bug was Pip interrupting itself. On a tablet speaker, its own voice came back through the microphone and the system heard it as the child speaking. For an adult that is an annoyance. For a child who needs a calm, predictable voice, it breaks the session. I now close the microphone while Pip speaks and reopen it a moment after Pip stops. Small details like that decide whether a parent keeps using it.
