This is a submission for the Presage SmartSpectra SDK Challenge

What I Built

We have all sent a message we regretted: to a client, a teammate, the family group. The words felt right in the moment because we were not.

Cool Off is a Flutter app (Android + iOS) that adds a short, honest pause before you hit send. You write or paste the message in the app, tap Check & Send, and a 20-second front-camera check reads your pulse and facial expression while the screen asks you to reread what you wrote. You get one of three results:

  • 🟢 Calm: looks good to send.
  • 🟡 Warm: maybe reread it.
  • 🔴 Heated: hold it for a few minutes, or take a one-minute cool-down.

Then you choose:

  • Send opens the normal share sheet (WhatsApp, SMS, email...). Cool Off never sends anything itself and never reads other apps.
  • Hold parks the message and sends a notification when the hold ends: "Still want to send it?"
  • Cool down is a minute with a breathing guide and a live pulse graph, then an automatic re-check and a before/after card (in my demo: 100 → 94 bpm, expression heat 2 → 0).
  • Send anyway is on every result. Cool Off never blocks you.

If you do send something while Heated, the next morning it asks "Glad you sent it?" 👍 / 👎, right from the notification or in the app. Those answers feed Insights (messages cooled off, regretted heated sends, average pulse drop after a cool-down) and slowly tune how early Cool Off speaks up.

This entry has two parts

To build Cool Off in Flutter I first needed SmartSpectra in Flutter, and there was no plugin. So the entry is:

  1. smartspectra_flutter, an open-source Flutter plugin for the SmartSpectra SDK, published on pub.dev (Android + iOS, 160/160 pub points). It wraps the official Android SDK and the Swift SDK behind one Dart API, so any Flutter developer can now use SmartSpectra.
  2. Cool Off, the app built on top of it.

Demo

The video is real screen recordings on a Samsung Galaxy S22 Ultra with real live readings. A few demo shortcuts are labelled on screen: the Send-anyway and cool-down scenes start from the reading of the Heated check shown just before them, Insights includes two weeks of sample history, and the next-morning check-in is shown right away instead of the next morning.

Code

Both are MIT licensed. Setup is in each README: a Presage API key in a git-ignored secrets.json, your own Firebase project (flutterfire configure), then flutter run. A physical device is needed for the camera; QA builds can also simulate a reading on an emulator.

How It Responds to Human Signals

Live signals used: pulse (cardio metrics) and facial expressions (anger, contempt, disgust).

1. A personal baseline. On first launch a calm 30-second reading stores your usual pulse and your resting face. Every check compares you with yourself, not with an average person, so a naturally high resting pulse or a face that "reads angry" at rest doesn't count against you.

2. A 20-second check, live. While you reread the message, the camera orb fills, a heart beats at your measured pulse, and the SDK's imaging guidance shows hints like "Too dark" or "Face not centered". The first pulse value arrives after about 12 seconds, which is why the check is 20.

3. Scoring. Three signals, each worth 0–2 points:

Signal Source 1 point 2 points
Pulse above your usual SmartSpectra pulse 5–15% over 15%
Anger + contempt + disgust above your resting face SmartSpectra expressions medium high
Text heat (CAPS, !!!, insults, dismissive phrases) on-device rules, no AI some a lot

Calm is 0–1, Warm 2–3, Heated 4+. Words alone can only reach Warm: Heated needs your body to agree. If the pulse can't be read, the result says so and scores the other signals; it never invents a value. The result screen lists the reasons in plain words ("Pulse +10 above your usual · lots of CAPS").

4. Cool-down, live. The cool-down screen keeps the camera running for a minute, plots your pulse against a dashed "your usual" line, and slowly shifts the whole screen from the result's colour to ice as you settle. Then it re-checks automatically and shows the before/after numbers.

5. It learns. Next-morning answers adjust the Heated cutoff (mostly regrets → it speaks up sooner, mostly "glad I sent it" → later), and sensitivity can be set by hand.

How I integrated the SDK: the plugin

smartspectra_flutter wraps com.presagetech:smartspectra:3.4.0 (Kotlin bridge) and SmartSpectra-Swift 3.4.0 (Swift bridge, Swift Package Manager) behind one Dart API, with features kept identical on both platforms:

final spectra = SmartSpectra.instance; await spectra.configure(SmartSpectraConfig( apiKey: apiKey, metrics: {SmartSpectraMetric.cardio, SmartSpectraMetric.expressions}, dnsFallback: true, )); spectra.metrics.listen((m) { final bpm = m.pulseRate?.value; // live pulse final mood = m.expression?.dominant; // Expression.angry, ... }); await spectra.start(); // with SmartSpectraPreview on screen 
  • A native camera preview widget (SmartSpectraPreview) with an onReady callback, so the app only starts measuring once the preview is on screen and the camera never runs unseen.
  • Streams for metrics (pulse, breathing, HRV, expressions, blinks), status, imaging guidance and errors.
  • The camera stops automatically when the app goes to the background, on both platforms.
  • dnsFallback: some home routers can't resolve Presage's servers. The plugin can route the SDK's traffic through a small loopback proxy that falls back to DNS-over-HTTPS, with TLS still end to end. Found and fixed while testing on my own Wi-Fi.
  • Tested end to end on a Galaxy S22 and an iPhone (iOS 27). 45 plugin tests; Cool Off has 114 (scoring, text heat, auto-tune, database, screen logic).

Prize Categories

I'm entering both categories.

Best Thoughtful Real-World Use. Angry messages are an everyday, universal problem, for anyone who has ever regretted a text. Cool Off fits where the problem happens: right before sending, in about 20 seconds, with no account. It is designed to help, not to police: Send anyway is always one tap away, results explain themselves in plain words, every check is relative to your own baseline, and the next-morning check-in lets you decide whether it was right. Accessibility: screen-reader labels on gauges and charts, text up to 2×, reduced motion respected. And because a good idea is only useful if others can build on it, the Flutter plugin is open source and on pub.dev.

Most Creative or Delightful Use. Most camera-vitals demos are stress meters. Cool Off uses the same signals as a pause button for your temper: a temperature theme from ember to ice, a heart that beats at your measured pulse, a cool-down screen that visibly cools as your pulse settles, and a before/after card that shows you actually calmed down. The next-morning "Glad you sent it?" turns your own regret into the thing that tunes the app.

Consent and Data Handling

  • Consent first. Before any camera use, the welcome screens explain what the camera does and what is stored, then ask for camera permission with an explanation. Nothing is measured before that.
  • Camera only during a check (calibration, check, cool-down), always with a visible preview. It stops the moment the app goes to the background (enforced by the plugin). No covert or background monitoring, and Cool Off never reads or intercepts other apps.
  • Video stays on the phone. Frames are processed on the device by the SmartSpectra SDK; they are never recorded, stored or uploaded.
  • Message text never leaves the phone: not to analytics, crash reports, the SDK or any server. It is stored only while a message is held, and kept afterwards only if you turn on Keep message text.
  • Storage and retention: baseline, check results, held messages and check-in answers live in an on-device database in the app's private storage until you delete them. No accounts, no backend, no sync. Settings → Delete all my data wipes everything and returns to the welcome screen.
  • Sharing: none. Sending is done by you, through the system share sheet.
  • SmartSpectra SDK: contacts Presage only to validate the API key and load its models. SDK telemetry is off, and LLM Insights / any AI service is not used.
  • Firebase Crashlytics and Analytics are on by default and can each be turned off in Settings. They carry only crash reports and a fixed list of usage events, never message text, recipients, pulse values, expression scores or results. The advertising ID is not collected.
  • Not a medical device. Cool Off is for wellness and reflection only. It makes no diagnosis or treatment claims and makes no medical decisions.

Thanks to Presage and the DEV team for the challenge. 💙