Personal projectFull-stack product2026Live

Meetora

Peer-to-peer video meetings with live captions, reactions, screen sharing, host controls and session recaps.

automated security checks
26automated security checks
WebRTC mesh with STUN + optional TURN
P2PWebRTC mesh with STUN + optional TURN
captions and searchable transcripts
Livecaptions and searchable transcripts
01Overview

Video calls in the browser that feel like a room: a pre-join lobby, an adaptive grid with active-speaker glow, live captions, reactions and raised hands, real host controls, and a searchable recap of every session.

OriginMeetora 2.0 is a full redesign and rebuild of a basic course project (“MeetUp”).

02Why I built it

The original app let anyone end anyone's meeting, because the server trusted whatever user id the browser sent. That one bug told me the real job: not more buttons, but a server that knows who is who, and a call UI that feels calm.

Questions I had to answer
  • Q1

    How do I prove who is on a socket, not just believe them?

  • Q2

    How does the UI know who is speaking without a media server?

  • Q3

    How do I share a screen without renegotiating every peer connection?

  • Q4

    What should people get after the call ends?

03Before → after
  • JoiningBeforeStraight into the callAfterLobby with camera preview, mic level meter, device pickers and who's already inside
  • In the callBeforeGrid, mute, camera, chatAfterAdaptive grid + spotlight/pin, active-speaker glow, screen sharing, reactions, raise hand, live captions
  • HostBeforeAnyone could end the meetingAfterHost-only end for all, ask to mute / mute all, remove participant
  • HistoryBeforeCard list with chat logAfterAttendance timeline, searchable transcript with export, copyable summary
  • SecurityBeforeSockets trusted the client's user idAfterEvery socket verifies a Clerk session token; host checks and room scoping on the server
04Key decisions
  1. Verify every socket with a Clerk session token

    Host actions and room membership are checked on the server, so the browser can't promote itself to host.

  2. Active speaker from real audio levels

    WebRTC receiver audioLevel for others, Web Audio for your own mic: speaker glow without an SFU.

  3. Screen sharing with replaceTrack

    Swapping the video track means nothing renegotiates, so sharing starts instantly for everyone.

  4. Captions through a voice gate

    An AudioWorklet and a voice gate send only speech to Gemini transcription on the server, with the Web Speech API as a fallback.

05Screens
Meeting grid with active-speaker glow
Pre-join lobby with camera preview and mic meter
Spotlight layout while presenting
In-call chat panel
Dashboard with activity chart and upcoming meetings
Session recap with summary
Attendance timeline for a session
Post-call screen
A call on a phone
Dashboard on a phone
06Features
Meetings
  • Instant meetings with readable codes, or schedule for later
  • Invite links and Google Calendar links
  • Pre-join lobby with device test
In the call
  • Adaptive grid, spotlight and pinning; presenters spotlighted automatically
  • Screen sharing (tab, window or screen)
  • Emoji reactions, raise hand, chat, people panel, keyboard shortcuts
  • Live captions
After the call
  • Day-grouped history with search and filters
  • Attendance timeline
  • Searchable transcript with .txt export and a summary
07Under the hood
01
Authenticated signalling

Socket.io signalling where every connection verifies a Clerk session token.

02
26 security checks

Automated security checks run against the server.

03
No-renegotiation sharing

replaceTrack swaps camera and screen without new offers and answers.

04
Bento dashboard

Stats, 7-day activity chart, up next, upcoming and recent sessions, plan usage.

08Tech stack
React 19ViteWebRTCSocket.ioClerkNeon PostgresGeminiTailwind CSS v4
09Looking back
The hard part

Permission edge cases. Camera blocked, mic missing, device in use: each one needs a clear message, and you should still be able to join to listen and chat.

What I learned

In real-time apps the server must be the source of truth for identity and roles. The UI can be optimistic; authority can't.