AI Engineer | Model Deployment | Fine-tuning | Inference
Your first fine-tune should teach you the whole loop.
I walk through it with Unsloth: data → baseline → training → evaluation → local model → Python.
Including a mistake I made—and why a completed training run doesn't mean you have a better model.
Full tutorial below 👇
https://youtu.be/YxXBojdD1as
UGC Video Prompt Automation System for Ad SaaS
Designed and implemented a prompt-generation system that converts structured ad scripts into best-in-class, scene-level video generation prompts for UGC-style TikTok/Reels ads. The workflow ingests Agent 2 outputs (concept, scenes, hooks, body, CTA, product positioning) and applies reusable prompt templates and logic to enforce character consistency, natural speech patterns, realistic lighting/camera imperfections, and casual, native social media aesthetics. Delivered a production-ready specification for Agent 3 integration, a reusable prompt library, and a fully generated UGC-style video produced end-to-end with this system.
Built a little Mac app called Flip, and it somehow went viral on Twitter.
The idea is pretty simple. What if every window on your Mac had a back side where you could keep notes? So I built exactly that.
Flip any window, jot something down, flip back. Your notes stay attached to that window.
No extra apps to switch between.
Built natively for macOS, completely local, and free.
Check it out: https://flip.pylondev.com
Here’s the original tweet:
https://x.com/skimmycat20/status/2096580098506084400