This is a submission for the Presage SmartSpectra SDK Challenge
What I Built
Hot Seat is a desktop interview coach that rehearses the nerves, not just the answer.
You answer interview questions out loud in front of your laptop. The webcam reads your pulse with the Presage SmartSpectra SDK, and Hot Seat:
- shows which question actually got to you, measured against your own resting pulse;
- offers a short breathing reset when your pulse stays high, and ends it by itself when you settle;
- ranks the session by "heat" and lets you practise the hottest questions first next time.
Everyone rehearses the answer. Nobody rehearses being nervous. The question that throws you is rarely the one you expect, and you usually find out in the real interview. Hot Seat finds it in practice, with a number attached.
From my own recorded session (live camera, real numbers, the same one you see in the video):
- resting pulse: 66 bpm;
- "Why do you want to work here?" peaked at +23 bpm and triggered a breathing pause that settled in 12 s;
- hottest question on average: "What are your salary expectations?" (+11 bpm).
The flow: consent → 30-second calibration → for each question, 15 seconds of think time, then answer out loud → results ranked by heat → Practise hottest first.
Demo
A 90-second cut of one real session with the live camera. Calibration and five of the seven questions are sped up, and the video says so on screen.
Code
arlanimanbayev
/
hot-seat
Practise the interview questions that make your heart race
Hot Seat
Practise the interview questions that make your heart race.
You answer questions out loud in front of your laptop. Hot Seat reads your pulse and breathing from the webcam with the Presage SmartSpectra SDK, shows which question got to you, offers a short breathing pause when your body says you need one, and puts your hardest questions first the next time you practise.
A real session with the live camera: resting pulse 66 bpm. "Why do you want to work here?" peaked at +23 bpm while thinking, triggered a breathing pause that settled in 12 seconds, and "What are your salary expectations?" came out hottest at +11 bpm on average.
Why
Everyone rehearses the answer. Nobody rehearses being nervous. The question that throws you ("What are your salary expectations?") is rarely the one you expect, and you only find out in the real interview. Hot Seat finds…
Setup: Node.js 20+ and a Presage API key, then npm install and npm start. If you have no key or no camera, a clearly labelled Simulated mode walks through the whole flow. npm test runs 14 tests for the session logic and the metric decoding.
How It Responds to Human Signals
1. Signals used
| Presage signal | Role in Hot Seat |
|---|---|
| Pulse rate | The core signal: baseline, live meter, heat score, pause trigger, recovery |
| Breathing rate | Context and a second pause trigger, only while you are silent |
| Talking | Shows when you speak; gates the breathing trigger; talking share per question |
| Expressions | Average "happy" score per question, in the exported numbers |
| Validation status | Plain-language framing and lighting hints, always on screen |
2. Live response
- Calibration (30 s). The median of your pulse readings becomes your resting baseline. Every number after that is relative to you, not to a population average.
- Think time (15 s per question). It works like the prep time in a one-way video interview. Presage's pulse rate is a 12-second average, so every question gets one full window, and the first hit of a question shows up before you have finished your first sentence.
- Live meters. You see pulse and breathing with the change from your baseline, a heat bar, and a sparkline of your pulse for the current question. A value older than 8 seconds is dimmed, so a held number never pretends to be live.
- Quick reset. If your pulse stays 12 bpm above resting for 8 seconds (or breathing 4 /min above resting for 8 seconds while you are silent), Hot Seat offers a breathing pause: a 4-in / 6-out pacer and one sentence that says what it saw ("Your pulse has been about 10 bpm above your resting rate").
- Recovery. When your pulse is back within 5 bpm of resting (and breathing within 2 /min), the pause ends by itself and the recovery time is logged ("settled in 12 s").
- Results. Heat = average pulse above resting, in bpm. Questions are ranked, the hottest goes in the headline, and Practise hottest first restarts with the same baseline and the hardest questions first. If nothing rose above resting, the results say so instead of naming a "hottest" question.
3. Integration
Hot Seat is an Electron app with a sandboxed renderer (contextIsolation, strict CSP, popups and navigation blocked). The camera permission is granted only for video, and only to the app's own page.
-
Main process:
bindSmartSpectraIpc(win)from@smartspectra/node-sdk/main. -
Renderer:
SmartSpectraSDKfrom@smartspectra/node-sdk/renderer, withstreamAvailablefor the preview,validationStatusfor the hints andmetricsfor the data. -
Decoding:
decodeMetricsturns each protobuf update into a reading.readingFromMetricsis a pure, tested function. -
Logic:
HotSeatSessionis a pure class with no DOM and no SDK. It handles the baseline, pause trigger, recovery and ranking, and it is unit-tested.
const sdk = new SmartSpectraSDK({
apiKey,
// Only what Hot Seat uses: face and upper chest are enough.
requestedMetrics: [M.BREATHING_RATE, M.PULSE_RATE, M.TALKING, M.EXPRESSIONS],
enableAccumulatedOutput: false,
enableTelemetry: false, // privacy first
});
sdk.on('metrics', (buf) => onReading(readingFromMetrics(decodeMetrics(buf)))); // simplified
4. What the signals taught me
- Ask only for what you use. My first build requested the full breathing bundle, which includes the abdominal waveform, and that needs the waistline in view. On a laptop camera the SDK kept saying the chest was not visible, and pulse barely came through. Requesting only pulse rate, breathing rate, talking and expressions fixed it: in the final session the hint stayed on "Good signal" and pulse updated continuously.
- Updates arrive split. Cardio, breathing and face data come as separate events. Hot Seat keeps the last known value of each signal. Otherwise the meters blink to "–" and a face-only update resets the 8-second pause timer. There is a test for exactly this.
- Respect the windows. Confidence is 0 until a window fills, so those placeholder values are ignored. Values outside Presage's validated ranges (pulse 40–110 bpm, breathing 5–40 /min) are dropped too. An early test showed a "breathing rate" of 53 /min while I was talking, which is exactly why.
- Breathing and speech don't mix. Breathing rate is a 30-second average and speech disturbs it, so it never counts toward the score. It only triggers a pause while you are silent: going quiet with racing breath is the freeze the pause is for.
- One clock. Questions, pauses and readings all use the app's own monotonic clock.
5. Limits
- One session from one person tells you what to practise, not a fact about you.
- Pulse needs light on your face and your head and upper chest in view. The hint says when that is not the case.
- These are estimates for practice feedback, not health information.
Prize Categories
Best Thoughtful Real-World Use.
- The need: interview anxiety is common, and practice tools only rehearse the words.
- The users: job seekers, students heading into internship interviews, founders before investor Q&A, anyone about to say a salary number out loud.
-
How the design supports them:
- no account and no server;
- one honest consent screen;
- hints in plain words, not error codes;
- think time modelled on real one-way video interviews;
- three decks (job interview, salary negotiation, startup pitch Q&A) or your own questions;
- an outcome you can act on: which question to drill next.
Most Creative or Delightful Use. Your own pulse becomes your practice plan. The app never judges your answer; it notices your body and turns that into the order you practise in, with a calm, self-ending breathing reset in between.
Consent and Data Handling
- Consent. Before anything starts, the setup screen says what the camera is used for and what is kept. It requires an explicit tick. macOS asks once for camera permission. The camera runs only during a session, and ending the session releases it.
- Storage. No video or images are stored, by Hot Seat or anywhere else. Pulse, breathing, talking and smile values are kept in memory for the current session only. Nothing is written to disk unless you press Export numbers (JSON), which saves a file where you choose.
- Sharing. Hot Seat has no server, no analytics and no account, and it uploads nothing. Measurement is performed by the Presage SmartSpectra SDK with your own API key, which authenticates with Presage.
- Retention and deletion. Hot Seat keeps nothing after the app closes. Delete this session clears everything from memory at once.
-
SDK telemetry. Hot Seat sets
enableTelemetry: false. Presage documents that its telemetry never includes raw video frames or measured physiological values (Telemetry and privacy). - API key. The key is kept in memory by default. It is stored in the app's local storage on this computer only if you tick "Remember".
- AI services. None. Hot Seat sends nothing to any AI model.
- Not medical. Hot Seat makes no diagnosis or medical claim. It is for practising on yourself, with your own consent, and not for monitoring other people.
Credits
- Built solo with the Presage SmartSpectra Node.js SDK, Electron, plain HTML/CSS/JS, esbuild and
node:test. - An AI coding assistant (Claude) helped write the code, tests and documentation. I made the product decisions and reviewed the thresholds and copy.
- MIT licensed.

Top comments (0)