I gave AI varying levels of control across 3 real projects for a month. One worked. One hit a hardware wall. One turned into a codebase I couldn't even read. Honest numbers included.
Field notes from a hands-on AI builder.
Learning AI, machine learning, and computer vision by building and breaking things. Real logs, budget GPU benchmarks, and honest experiments.
Latest Notes
View all →I used Google AI Studio’s free Gemini 3.5 Flash to build a full travel itinerary app, handling everything from UI design to database integration. Here’s what worked, what broke, and whether you can actually launch with just free tools.
I built a complete AI dev team — PM, Developer, QA agents. All local, no cloud, no API costs. Here's the honest result.
I ran a small local AI model on a budget 4GB GPU to see if it can handle real tasks. Here’s what actually worked and what didn’t.
I ran five local Large Language Models (LLMs) on an outdated Nvidia GTX 650 with only 4GB of VRAM. Here’s what actually worked and what didn’t.