My Coffee Stain AI: How I Built a Local LLM That Reads My Spills (No Cloud Required)
So, I've been making coffee at my desk for years, and my notes? They're a disaster. Spilled coffee = instant text chaos. One Tuesday, I spilled a latte right on my meeting notes, and instead of cursing, I thought: 'What if my laptop could read this mess?' Turns out, it can - and without needing Google to scan my desk. I used a tiny, free local LLM called Llama.cpp (runs on my 2019 MacBook!), trained it on 50 photos of my 'stain patterns' labeled as 'meeting', 'deadline', 'coffee break', etc. Now, when I snap a pic of a stain, it instantly translates it to text like '2pm meeting' - no internet, no cloud, just my laptop. It's not magic; it's just tiny data + local AI. Why does this work? Because local LLMs are way smaller than you think. I used a 8MB model - way smaller than ChatGPT. No privacy risks, no data leaving my machine. And it's perfect for my chaotic desk: the 'coffee ring' next to my calendar? It's...