SyncItUp

Inspiration

Swiping rewards how you look, punishes everyone who doesn't photograph well, and leaves the rest of us feeling judged. We asked ourselves what happens if nobody swipes at all, and the app gets to know who you are (how you think, talk, react and what you love) before anyone sees what you look like. Our motto became: No Swiping, just Vibin'.

What it does

SyncItUp replaces matching with syncing. There's no like button and no rejection screen.

• Onboarding without boxes: name, pronouns, birthday, height and even eye colour (don't ask why). Gender is picked on a colour spectrum, not a checkbox, and you stretch a window over the same spectrum for the genders you're attracted to, or go full rainbow 🌈.
• Virtues and bio: job, degree, gym habits ("Nah, I'm perfect as-is" is a valid answer), languages, and a 400-character bio with conversation starters.
• Full body scan: real-time pose detection guides you through a 360° turn. It isn't about looks; it's about how you carry yourself.
• The vibe check: a live voice interview with an AI interviewer. While you talk, the app shows a multimodal dashboard: face-emotion detection from your camera, a live mic waveform, a heart-rate monitor and an 8-channel EEG stream. At the end you get your vibe: Late-night Philosopher, Chaotic Good, Main Character…
• Syncs, not matches: the algorithm decides who you're on the same frequency with, and the chat is simply open. Every sync shows why you synced ("Your EEGs peaked on the same answer. Spooky.").
• Sync map: a UMAP-style 2D map of every user, clustered by vibe, showing where you sit and lines to the people you're synced with.
• Activities that sharpen your syncs: talk to the AI again, answer written questions, Sync by reel (doom scrolling becomes sync scrolling: the more you watch, the more you sync with people who share your taste), Moral Mirror (timed dilemmas from a slightly-off future that measure how long you hesitate), and Memory Auction (100 coins, and you only keep the memories you bid on).

How we built it

• Next.js and React with plain CSS: a mobile-first web app with a bold pink-on-light design system and no UI libraries.
• In-browser AI: face-api.js for facial expression recognition, and TensorFlow.js with MoveNet for pose detection and tracking the 360° turn.
• Voice: the browser's speech synthesis for the interviewer, plus a silent recorder with voice-activity detection that ends your turn when you stop talking. All of it sits behind a swappable voice layer, ready for a local speech-to-speech model.
• Simulated sensors: an EEG generator with realistic alpha/beta/theta waves and eye-blink artifacts that react to whether you're speaking, and a smoothly drifting heart rate.
• The sync map: a seeded 2D embedding drawn as SVG, with an animated convergence and a colour palette validated for colour-blind safety.
• Deployment: a static site on GitHub Pages. The mock backend runs entirely in the browser with localStorage, so there are no servers to run and every visitor gets their own data.

Challenges we ran into

• Browser speech recognition beeped and opened system popups in some browsers, which ruined the immersion. We replaced it with our own silent mic recorder and voice-activity detection.
• Three people on one interview page: voice, vision and UI landed in parallel, so we had to keep each part separate (conversation logic, voice engine, camera panel) to avoid stepping on each other.
• Heavy libraries: fitting the ML models into a phone-sized web app meant handling double-mounted effects, cameras that never got released, and libraries that tried to load Node-only code in the browser.
• Going static: moving from server API routes to a fully client-side backend so the app could live on GitHub Pages without losing any features.
• Making fake data feel real: realistic EEG traces, a heart rate without steep jumps, and a UMAP plot whose labels never collide.

Accomplishments that we're proud of

• A complete, clickable product from landing page to live chat in one hackathon.
• Real computer vision (emotions and full-body pose) running on a phone in the browser.
• A gender and attraction picker that treats identity as a spectrum, not a dropdown.
• A dating app where nobody can be swiped away, and the design makes that feel like a feature, not a gimmick.
• The vibe check dashboard: an AI interviewer, EEG, heart rate and emotion detection on one screen is genuinely fun to watch.

What we learned

We learnt that dating apps are not about what a person wants, but what is the right match for you: sometimes a person doesn't even know (or understand) what the perfect match for them would be.

What's next for SyncItUp

• Real speech-to-speech: a local voice model so the interviewer listens, understands and asks follow-ups.
• Real embeddings: multimodal models turning voice, text, taste and the body scan into the vectors behind the sync map and sync scores.
• Wearables: live heart rate from smartwatches, and real EEG headbands for the brave.
• Twin Date and Prediction Check: your AI twin goes on a simulated first date with a sync's twin, and you rate how well it knows you.
• Activities IRL: group plans built around shared vibes, so syncs turn into real meetings.
• Privacy by design: keeping as much processing on-device as possible, because the more an app knows about you, the more carefully it has to treat it.

Built With

• javascript
• react