Inspiration
Everyday routines rarely run from start to finish. A phone call, a delivery or a family member needing help breaks the flow. The hard part is not stopping, it is starting again: Where was I? What did I already do? What is the very next thing? Most task tools assume a clean start. I wanted to explore what an Alexa+ moment for "where was I?" could feel like.
What it does
NextCue is a simulated Alexa+ experience in a web app (the Alexa+ track's simulation path). It holds your place in an everyday routine:
- One clear next step instead of a long to-do list (built-in routines: study session, pack a bag, tidy your desk).
- Pause with a cue: when you are interrupted, leave a short note such as "Phone call; notebook open on the desk".
- Resume exactly: NextCue shows the unfinished step, your note, the last completed action and how long you were away. Nothing is skipped or lost.
- Make this step smaller: built-in steps have short, hand-written microsteps. Finishing a small step never skips the rest of its parent action.
- Cue panel: typed commands simulate the Alexa+ conversation:
start study,done,pause <note>,resume,smaller,status. Unknown commands only return help and never change progress. - Custom routines with one action per line, plus an optional focus timer and read-aloud using the browser's own voice.
- Privacy by default: no backend, account, analytics or microphone. State lives in page memory unless you turn on "Remember on this device" (localStorage).
How I built it
- Plain JavaScript ES modules, HTML and CSS, with no build step and no dependencies.
- A pure, immutable state engine (
engine.mjs) handles routines, pause/resume, microsteps and command parsing. The UI (app.js) only renders state and dispatches actions. - Nine automated tests with Node's built-in test runner cover interruption and resume, explicit completion, smaller-step progress, state restoration, invalid commands and immutable transitions.
- Accessibility: labelled controls, a polite live region for the conversation, focus management after every action and a skip link.
- AI assistance: OpenAI ChatGPT/Codex assisted design, implementation and testing of the app. Claude helped produce the demo video (synthetic narration) and the submission materials.
Challenges I ran into
- Making resume trustworthy. Resuming must never skip, duplicate or lose a step, even when a microstep is in progress or a command is invalid. Keeping the engine pure made these cases testable.
- Smaller without skipping. Microsteps had to feel like progress while keeping the parent action open until it is really done.
- Honest scope. It is tempting to fake "AI". NextCue uses deterministic local rules and says so in the UI.
Accomplishments that I'm proud of
A calm, complete experience that works offline from a static folder, with an engine covered by tests and a UI that clearly says what it is and what it is not.
What I learned
Interruption recovery is mostly about context: the last completed action, a human note and the very next step matter more than reminders or streaks.
What's next for NextCue
- A self-hosted MCP server (spec 2025-11-25, Streamable HTTP) exposing
start_routine,complete_step,pause_with_note,resumeandget_status, so a real Alexa+ conversation can drive the same engine. - Usability testing with people whose days are full of interruptions.
- Optional, explicitly scoped cross-device storage.
NextCue is a prototype. It is not connected to Alexa or Amazon, makes no clinical or productivity claims and is not a medical product.
Log in or sign up for Devpost to join the conversation.