Project Details
The problem, and why this exists
Most ADHD tooling is a phone app: another notification competing with every other notification, easy to dismiss, easy to ignore, and disconnected from the physical environment where the actual hard moments happen: getting out of bed, winding down at night, noticing a day has drifted without structure. A notification is a request for attention you already don’t have enough of. A light turning on, or the house itself doing something different, doesn’t ask for attention the same way. It’s just true, and it acts in the environment rather than competing for a slice of it.
This system exists because that distinction mattered enough to build real infrastructure around it, for personal use, over an existing smart-home setup rather than a phone app store.
What was built
- A graduated wake-escalation ladder: lights, then audio, then stronger prompts, each stage triggering only if the previous one didn’t work, rather than one alarm that’s easy to silence half-asleep and forget.
- Calendar- and weather-aware activity nudges, so a prompt reflects what’s actually coming up that day rather than a generic reminder with no context.
- A dynamic wind-down and screen-curfew check that adapts to the actual evening rather than firing at a fixed clock time regardless of context.
- A weekly digest tracking sleep and consistency trends over time, so the system reflects back a pattern instead of only ever reacting moment to moment.
Technology choices, and why
Built into existing home automation, not a standalone app. The escalation ladder and nudges act through devices already in the room (lights, speakers), specifically because that’s not something to open, check, or dismiss; it’s just what the house is doing. The cost is real infrastructure work (home-automation platform integration, timers, device orchestration) instead of shipping a simple app; the benefit is a support system that doesn’t depend on remembering to open anything at the exact moment it matters least.
A three-provider text-generation fallback chain behind every generated nudge: a local model tried first, then two independent cloud providers, then a plain static message as the last resort. A system meant to help at the hardest moments of the day (waking up, winding down) cannot be allowed to go silent because one AI provider is having a bad night; the fallback chain exists specifically so “no message at all” is nearly impossible, not just unlikely.
Trend tracking over point-in-time reminders. A single day’s nudge is useful in the moment; a system that also shows the pattern across weeks is what actually helps someone notice drift before it becomes a real problem, which is the harder and more valuable thing to build.
Outcome
A live, always-on system in real daily personal use, not a prototype or a concept. The escalation ladder runs against real smart-home devices every morning; nudges are calendar- and weather-aware in practice, not just in design; and the text-generation fallback chain has been genuinely exercised by real provider outages, not just designed for a hypothetical one.
This is also, deliberately, the most personal project in this collection, built to solve a real problem for its own builder first, with the same engineering discipline (real fallbacks, real monitoring, no silent failures) applied to personal infrastructure as to client-facing systems.