rajeevn

The bottleneck for power users isn't compute or model capability anymore—it's our own throughput in processing information, applying judgment, and crafting prompts. This post looks at converting continuous computer use data into structured events to train lightweight personal models, exploring whether local scaling laws apply to continual personal fine-tuning. Curious if others are experimenting with local fine-tunes on their own workflow telemetry.

Was flutter considered a candidate for the technical solution.