The live address and real content stay private on purpose. The screens show the real UI, run locally with sample data (sample site name and titles, my own AI videos).
← All work
Case 05 / 09
AI Video Streaming Platform
A fully serverless video streaming service on the Cloudflare edge — AI auto-translation into 10 languages and crypto subscription payments, built solo with AI
Live
2026.09 – 2026.10 (~3 weeks, from file dates — no git) / Solo (AI as the team) / Web (serverless)
- Problem
- Could one person run global video streaming — playback, subscriptions, many languages — with no servers?
- My role
- Planning · setting the bar · directing AI to build · deployment
- Stack
- Cloudflare Workers · Hono · D1 · R2 · Workers Rate Limiting
Results
- Running on Cloudflare Workers — no servers to manage
- Live
- Running on Cloudflare Workers — no servers to manage
- Multilingual UI with AI auto-translated video titles
- 10 languages
- Multilingual UI with AI auto-translated video titles
- Subscription payments verified on-chain
- Crypto
- Subscription payments verified on-chain
- Lines of code, 39 DB migrations, ~70 versions
- ~11,300
- Lines of code, 39 DB migrations, ~70 versions
01
02
Idea / Problem
Could one person run a global video streaming service — fast playback, paid subscriptions, many languages — without operating a single server? The goal was to do everything at the edge (network points close to viewers), so there is nothing to patch, scale, or keep alive.
03
The Plan & Bar I Set
- No servers: compute, database, media storage, rate limiting, and AI translation all on Cloudflare.
- Payments must be verified, not trusted — a subscription activates only after the payment is confirmed independently.
- Every viewer sees the service in their language, including video titles, without manual translation work.
04
Directing AI
- Split the work into delivery, payments, localization, and security, and had AI build each against written rules.
- Kept a running README of what changed in every version and why — about 70 versions in ~3 weeks.
- Automated deploys with one idempotent script so a release is a single command.
05
Results
- Live on Cloudflare Workers + D1 (database) + R2 (media) + Workers Rate Limiting + Workers AI.
- 10 languages, with titles auto-translated by Workers AI.
- Crypto subscriptions: payments checked against the blockchain record and protected from reuse; provider webhooks signature-verified; one active device per subscription.
- Byte-range video streaming from object storage plus edge caching.
- ~11,300 lines of TypeScript, 39 database migrations.
06
Retrospective
The edge stack let one person launch a global service with zero server operations. What I'd do from day one next time: Git and tests before the first deploy, and monitoring with backups baked into the deploy script.
More screens · 6
The real UI, run locally with sample data — site name, titles, videos and wallet address are samples; the videos are my own AI work.
(01) Review
Honest Review
If a developer had built it
Strong4
Payments — verified, not trusted
Each crypto subscription payment is checked against the blockchain record and can't be reused; payment-provider webhooks are signature-verified.
Security basics — consistent
Signed admin sessions with strict cookies, login lockout, security headers, every SQL query parameterized (217 places), 6 rate-limit rules.
Performance — edge-first video delivery
Byte-range video streaming from object storage and edge caching, so no origin server is needed.
Speed & documentation — ~3 weeks, 70 versions, every change explained
~11,300 lines and 39 migrations; a 651-line README records why each version changed. Dates from file timestamps (no git), a minimum.
On par3
Architecture — standard 3-layer Workers app
Domain logic / views / routers split; a couple of large route files remain.
Release & ops — one-command deploy
A deploy script creates storage, runs migrations, deploys, and sets secrets idempotently — automated, but not CI.
Localization — 10 languages
10-language dictionaries plus AI auto-translated titles; accessibility at a basic level.
Next step (in progress)3
Testing & version control — no tests, no git
Backups were versioned ZIPs.
Plan → Private Git repo, then tests on payment verification and CI that also runs the existing type check.
Code consistency — a few 2,000-line files
5 files over 500 lines, some untyped spots, leftover temp files.
Plan → Split by domain and clean up as part of the move to Git.
Monitoring — little visibility in production
Minimal logging; no backup or rollback routine.
Plan → Add error tracking, scheduled DB backups, and a rollback step to the deploy script.
Next project →
Money Navigation
An app that helps travelers lose less money on every overseas payment — finished in a 48-hour hackathon by directing multiple AI sessions
