A Helper Small Enough for the Cheapest Phone

Many first-time smartphone users in Nepal get stuck on one small question: where do I tap? We are testing whether a tiny, offline helper can answer it by pointing at the right button and speaking the steps, without ever taking over. The first question is simple. On one low-end Android phone, how often does what the phone already knows about its own screen show the way, and when it doesn’t, how many of the gaps can a small vision model fill?

On one low-end Android phone, what fraction of UI-grounding intents across 5 Nepali apps and portals can be resolved from the accessibility tree alone, and how many of the failures can Florence-2-base recover?