A 7 TB recordings archive, rebuilt serverless
Hosting cost down about 66%, 17,000+ meetings searchable and gated
a loan-officer coaching company (700+ students, US) · 2025 to 2026
AWS S3CloudFrontLambdaPostgresReact
Problem
Years of coaching-call recordings lived in a cloud video account with rising fees and no way to search, gate, or reuse them. Deleting was not an option; the calls are the product's history.
What I built
- A pipeline that pulls every recording, transcript, chat, and summary into an S3 archive with clean metadata, about 7 TB.
- A portal on top: magic-link sign-in tied to the customer database, a searchable index of 17,000+ meetings, playback through signed CDN URLs so the video bytes never touch a server.
- Then the whole thing moved from a single always-on server to a serverless stack: static front end on a CDN, an API on Lambda, metadata in Postgres.
- Later fixes: a classifier that keeps webinars out of the deletion path, a metadata backfill that restored durations for 1,400+ recordings, and an admin transcript download.
Result
About 66% lower hosting cost, no single server to babysit, and a searchable archive that other systems (recap generation, quality review) now read from.
What it shows
Cost and reliability together: fewer moving parts, not more.