Skip to content
All work
Realtime · SaaS + Desktop2025

Loom-class Realtime Video SaaS + Desktop App

A Loom-style async video platform with a native desktop recorder, auto-summarisation, searchable transcripts and resumable chunked uploads.

LIVE · 4K
Realtime · SaaS + Desktop
Loom-class Realtime Video SaaS + Desktop App
38ms
Ingestion latency (median)
4.9M
Minutes recorded, Q1
+72%
MoM paid seat growth
0
Dropped uploads
The problem

The founding team had a browser-only MVP that dropped recordings on shaky connections and could not handle multi-monitor capture.

Transcription was a cron job. Summaries took 30 minutes. Revenue was gated by a painful onboarding.

They needed a paid-tier product inside eight weeks to stay on plan for a Series A raise.

The approach
  1. Shipped an Electron desktop recorder with camera + system audio + multi-monitor, using native capture APIs and a resumable S3 multipart uploader over WebSocket.

  2. Introduced an FFmpeg stitching pipeline on a small GPU pool; chunks are transcribed as they land.

  3. Streaming summaries over Server-Sent Events using a two-stage LLM chain: speaker-labelled outline → exec-ready brief.

  4. Rebuilt the dashboard in Next.js 15 with shared-element transitions, presence and realtime collab comments via Socket.io.

Architecture
Rendering architecture diagram…
Outcome
  • 38ms median ingestion latency, down from 6.2 seconds.

  • Zero dropped uploads across 4.9M recorded minutes in the first quarter.

  • 72% MoM growth on paid seats after the desktop launch.

  • Closed Series A three weeks ahead of plan.

What I owned end-to-end
  • System design across web + desktop + backend
  • LLM pipeline: transcription, outline, exec summary, eval harness
  • Observability: OpenTelemetry traces, Grafana dashboards, cost per minute
  • Hiring of two product engineers and the onboarding of AI QA
Next up

AI-Powered Webinar Platform with Voice Sales Agents