Glance

• 

  Facial Recognition with gesture based controls for networking events

Inspiration

Walking into a networking event is awkward. The room is full of strangers, and you have no idea who you should talk to. Most people end up hovering near the snacks or sticking with someone they already know. Glance fixes the start of the night by pointing you to people like you and giving you topics to open with, so the first hello isn't a guess.

The end of the night has its own problem. You leave with 20 new LinkedIn connections and almost no memory of what you talked about with each of them. Business cards and contact swaps save the name but lose the conversation, and the conversation was the valuable part. We wanted Glance to help you find the right people, start real conversations and remember them, while staying out of the way when you're actually talking.

What it does

Recognize: Point the LIVE Camera at someone who has also opted in, and Glance shows their name and a few talking points so you can skip the small talk.

Record: The conversation is recorded in the background with consent both people gave at the event entrance, so there's no extra approval screen.

Finish: When you're done, you each raise three fingers (thumb, pointer, middle). Each phone recognizes the other person's hand, flashes green and saves the connection.

Recap: You both get the same recap (topics, highlights and follow-up ideas) and a shared chat to keep talking.

Glasses: Wearing Ray-Ban Meta glasses, you can switch the conversation camera to the glasses, so the gesture happens naturally from your point of view.

How we built it

App: Expo and React Native with Expo Router. It runs as a web app on Vercel and is set up for a native iPhone build.

Backend: Convex handles events, pairing, real-time conversation state, hand-signal sync, uploads and recaps.

Face recognition: On-device with face-api. Face embeddings stay on the phone, and a short hold-to-confirm countdown prevents false matches.

Hand gestures: MediaPipe Hand Landmarker plus our own pose classifier. It checks that three fingers are extended and the ring and little fingers are folded, and requires the pose to stay stable across several frames.

Recaps: Meta's speech-to-text with speaker separation (diarization), followed by a Meta text model that writes the shared summary. Audio is deleted after it's summarized.

Challenges we ran into

Two-phone timing: Getting two phones to agree on when a conversation starts, who records, and when it ends. We pair face sightings that happen within 8 seconds of each other, and finishing works from either phone.

Gesture feel: The first version required recognizing both your own hand and the other person's hand on both ends. We changed it so each phone only looks for the other person's hand and holds for under a second.

Speed: Loading face models and opening the camera felt slow, so we added model preloading, camera handoff between screens, and a shorter connection screen.

Live vs. demo drift: The demo mode got ahead of the live, backend-connected version, and we had to bring recording and recaps back in line.

Accomplishments that we're proud of

The full loop works end to end in live mode across two phones: recognize, talk, three-finger finish, then a shared recap and chat.

Consent is handled once at the event entrance instead of with pop-ups during conversations.

The phone and glasses camera picker respects platform policy without hurting the experience.

What we learned

In social apps, every second of friction is visible to the other person. Shaving seconds off holds and loading screens mattered more than adding features.

Read platform policies early, because they shape the architecture, not just the launch checklist.

Use demo mode for showing the idea and live mode for proving it, and keep them in sync.

What's next for Glance

Test on real iPhones and Ray-Ban Meta glasses at an actual event.

Ship a native iPhone build with background audio and smoother camera handoff.

An organizer dashboard showing how many real connections the event produced.

Smarter follow-ups: reminders, calendar suggestions and recaps you can export to your CRM.

Apply for Meta approval to explore richer glasses features responsibly.

Built With

• convex
• expo.io
• face-api
• mediapipe
• meta-ai
• meta-wearables-dat
• react-native
• swift
• typescript
• vercel