Independent engineer · Cloudflare-native

AI systems that actually ship.

I design and build the whole chain: edge APIs, opaque-token relays, GPU render boxes that boot on demand and vanish when idle, live data pipelines and bots. No dashboards full of promises — working services, tested end to end.

Start a project See live work
vylux@edge — boot.log
$ vylux boot --profile production

Live work

shipped · running now

Live sports surfacesports · live

Live sports surface

Fixtures, scores, standings and top scorers from a normalised feed, plus an operator-gated HLS relay with expiring AES-GCM segment tokens. Own frontend, zero vendor names in the wire.

workershls relayaes-gcm98 tests
sports.vxlabs.us.ci
AI video studioai video · private

AI video studio

Text-to-video on an open-weights model running on a GPU box that wakes on request, registers itself, renders, and exits so the meter stops. Opaque job tokens, media relayed through the edge — the render host is never exposed.

gpu on-demandhmac tokenspytorchprivate beta
Invite only — ask for access
Render bridgeinfra · internal

Render bridge

Queue and job layer that sits between the edge API and a disposable GPU host. Signed registers, staleness windows, single-job queueing, media proxied so the origin host stays invisible to clients.

queueingkv registrysigninginternal
Powers the studio

What I build with

chosen for shipping, not trends

Cloudflare Workers Cloudflare Pages KV · D1 HMAC opaque tokens AES-GCM relays Node test suites PyTorch · diffusers On-demand T4 GPU HLS streaming Python automation WhatsApp cloud API Zero vendor leakage

Have something that needs to actually work?

Bots, edge APIs, AI pipelines, live data. Tell me the outcome you want and I'll tell you what's realistic — and what isn't.