Zayf Get in touch

Zain, software engineer

Software made carefully, shown in the open

Full-stack and AI engineer. I built and ran an AI video SaaS to about $10K MRR as its only engineer, and set up its partnership with BytePlus, part of ByteDance.

Now I build products for clients and for myself, and share how I build them: the tools, the agent setup behind them, and what breaks along the way.

Built
AI video SaaS, alone · ~$10K MRR
Partner
BytePlus · ByteDance
Client
Awards & Engraving Florida · Shopify

See the work Get in touch

01Work

What I've built

KreatorFlow

2025 – 2026 · CTO and sole engineer

An AI image and video SaaS that I built and ran alone as CTO: product, billing, cloud infrastructure and the generation pipeline. It reached about $10K MRR. I left in 2026 and kept the rights to the code I wrote.

  • ~$10Kmonthly recurring revenue
  • 1,278commits, one author
  • 406API routes
  • 76data models

While I was CTO, I set up KreatorFlow's official partnership with BytePlus, the enterprise AI arm of ByteDance, the company behind TikTok. I was on regular calls with their engineers and leads to take their Seedream image and Seedance video models into production. The partnership is still listed publicly

What shipped, month by month

  1. Month 2First generation feature: character images.
  2. Month 3Video, Stripe subscriptions and affiliates.
  3. Month 4–5Crypto checkout. Serverless GPUs on RunPod running ComfyUI.
  4. Month 7Voice, realtime updates, and BytePlus Seedream image models in production.
  5. Month 8Moved onto AWS: webhooks into Lambda, live results to the browser over AppSync.
  6. Month 8–9Seedance video, face swap, refunds and BytePlus cost tracking.
  7. Month 9Public developer API: keys, async jobs, signed webhooks, OpenAPI.
  8. Month 10–11Billing portal with proration, more video models, Discord bot, moderation.

Awards & Engraving Florida

2026 – now · Client · Shopify

A Shopify store for a Central Florida awards and engraving shop. I redesigned and rebuilt the whole site, every page of it, and keep adding to it.

ae-florida.com Trophy configurator Dog ornament

  • The whole siteEvery page redesigned and rebuilt
  • Catalogs hubRebuilt
  • GalleryRebuilt
  • Quote flowNew, with Meta Pixel and GA4 conversions
  • Trophy configuratorBacked by real product variants
  • Ornament personalizerOn-device ML, in the browser

Most recent: the dog-ornament personalizer

  1. PhotoThe customer uploads a photo of their dog
  2. CutoutOn-device ML cuts the dog out, right in the browser
  3. OrnamentTheir dog, on the ornament

The cutout runs on the customer's device, with no upload server.

02Projects

What I'm building

AnimeFlow

2026 · Desktop · Rust · TensorRT

A desktop anime player that upscales video with AI while it plays. Every frame goes through a TensorRT super-resolution model, then a chain of my own GLSL shaders, and comes out at 2880×1620.

Around the player: an AniList-linked library, subtitle search, frame-accurate clipping, and Watch Together, which streams the enhanced picture to friends over Tailscale.

  1. mpvDriven by a Rust sidecar
  2. TensorRTSuper-resolution model
  3. GLSL shadersDeband, line sharpening, colour
  4. 2880×1620On screen

Every frame has to land inside 41.7 ms at 24 fps, on one RTX 4060

26.8 ms per frame

Getting there took careful tuning: a TensorRT build tuned to its highest optimisation level, CUDA graphs with two inference streams, and a hold gate that reuses the last upscaled frame while a drawing is held still. In one episode I measured, that was a fifth of all frames.

UpscaledHeld, last frame reused

  • +62%from tuning
  • −22%AI work
  • 0dropped frames

Penwise

Now · Open source soon

A pen-first notebook for my Galaxy Tab S9 and its EMR pen. A Rust ink engine draws every stroke bit-identically on the tablet, in the browser and on my PC. Ink syncs live as I write, and AI agents can read, search and annotate my pages over MCP. Open source, with a demo, soon.

  • 169/169stroke goldens identical, native vs WebAssembly
  • 13brush presets
  • 14MCP tools

Milestones

  1. M0Toolchain
  2. M1Ink engine
  3. M2Editing engine
  4. M3Live sync + web view
  5. M4Tablet app
  6. M5Agent tools (MCP)
  7. M6Handwriting search

DoneIn progress

My agent setup

Daily · Linux

Behind all of this is a setup I've built up from everything I've shipped. I work across Claude Code, Codex, omp, Factory and pi, and I've built agent harnesses of my own. Around them I designed one system: about 100 skills, my own agent roles, rules, hooks and verification gates, and the knowledge I've gathered running real products, so work gets checked before it ships.

It's how I build client sites, Penwise and AnimeFlow. It also runs a gated pipeline for short motion ads, from brief and music through storyboard, build, QA and an independent critique. My logo, avatar and banners were one small job for it.

03What I build

Bring me an idea and I'll take it to a product people pay for, and keep it running.

Day to day

  • TypeScript
  • Rust
  • Python
  • SQL
  • Next.js
  • Postgres
  • AWS
  • RunPod
  • TensorRT
  • Stripe
  • Shopify
  • WebAssembly
  • Kotlin

04Path

How I got here

  1. The startSelf-taught

    No degree. I learned by building real products and shipping them.

  2. 2025 – 2026CTO, KreatorFlow

    Built and ran an AI SaaS to about $10K MRR, alone, and set up its BytePlus (ByteDance) partnership. Left in 2026 with the rights to my code.

  3. 2026 – nowClient work

    Redesigned and rebuilt a whole Shopify store, with a configurator and an in-browser ML personalizer, for a Florida awards shop.

  4. 2026 – nowZayf

    My own products, built in public: tools, videos and what I learn.

05Contact

Say hello

A product to build, an engineer to bring in, or a collaboration. I read everything.

hello@zayf.online