AI summary by TwinMind

We stopped using serverless. The results are insane.

Theo - t3․gg

18m 26s

Platform Statistics & V7 Launch Announcement

  • Wild year and a half since launch: 8+ million files uploaded
  • Hundreds of thousands of users, tens of thousands of applications
  • 6 months of development for complete overhaul (internal/external)
  • V7 launch: up to 5x faster upload speeds
  • Dog-fooding V7 with pic thing since launch

Key V7 Features & Improvements

  • Speed: Up to 5x faster uploads (especially for smaller files)
  • Reliability: Pause/resume functionality for uploads (S3 can't do this)
  • Framework Support: Added Remix support package
  • Legacy Infrastructure: Started shutting down original S3 direct upload system
  • Upgrade Process: Zero code changes required for many users - just package swap + environment variable change

Performance Benchmarks

  • Small files (800KB): 4 seconds → 0.5 seconds (377% faster)
  • Multi-image test (4x ~1MB files): 5 seconds → 1.5 seconds
  • Network conditions: 5ms latency San Francisco gigabit connection
  • Scaling: Performance now scales better with file size/count vs fixed overhead

Technical Architecture Changes

Old V6 Architecture (7 request hops):

  • User browser → Your server → UT API → S3 direct upload
  • S3 completion trigger → Lambda → UT API → Your server → User polling
  • Ghost files possible (security issue)
  • No upload validation/processing capabilities
  • Complex polling mechanism required

New V7 Architecture (3 request hops):

  • User browser → Your server (self-signed) → UT Ingest Server → S3
  • Connection held throughout upload
  • Direct response forwarding from your server back to user
  • Eliminated polling entirely

Infrastructure & Bundle Improvements

  • New Ingest Server: Custom Fastify-based server (no longer serverless)
  • Bundle Size Reduction: 30% reduction on top of previous 84KB cut
  • Final Bundle: 15.6KB minified + gzipped
  • Regional Deployment: Ingest servers in all UT operating regions
  • Token Structure: New upload thing token (replaces simple secret)
    • Combines: secret + app ID + region
    • Eliminates round-trip API calls for server selection

New Capabilities Unlocked

  • File Processing: Server-side file processing/validation
  • Security: Automatic file scanning for illegal content (future paid feature)
  • Compliance: Better path for HIPAA/SOC2 compliance
  • Bring Your Own Bucket: Now possible with new architecture
  • Self-Hosted: Future ability to deploy UT server in customer infrastructure
  • Resume Uploads: Spotty connection handling
  • Real-time Validation: File size, capacity, safety checks during upload

Team & Development

  • Key Contributors: Julius (infrastructure), Mark
  • Development Timeline: 6-month focused effort
  • Work Impact: Held up other development during this period
  • Relief: Team excited to ship after long development cycle

Business Direction

  • Enterprise Ready: No longer just for hobby projects
  • Sales Process: Schedule calls available for business customers
  • Future Features: Many new capabilities now unlocked by architecture change

TwinMind

Get summaries like this for your own meetings

TwinMind takes notes in the background during meetings, lectures, and calls, then writes the summary for you.

Free plan to get started

No bot joins your calls

Phone, desktop & Chrome apps

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome

Your next meeting, summarized like this

TwinMind listens in the background and turns meetings, lectures, and calls into searchable notes you can chat with. Start free.

Available on phone, desktop & Chrome