Mobile and AI engineer. I started with PHP and Django, spent years on Flutter apps and map software, and now work on language models, agents and automation.
Galli
Vector maps and turn-by-turn navigation, as two Flutter plugins. The gap between a line on a screen and someone actually finding their way.
Kipakosa
Point a phone at a printed photo and it starts to move. No app. It runs in the browser's camera, and every page of an album can play its own clip.
Nepali to English
A small translator with attention. Word order flips between the two languages, and attention is how the model follows it.
Lower loss, worse model
I fine-tuned Qwen2.5 to be more truthful on a single GPU, and had a local 7B model judge the runs. The run with the lowest training loss came last.
Agents, with a barrier
Chatbots and multi-agent systems wired into real work, always with a person's approval before anything irreversible.
A fleet you can read
Sales and machine data for a fleet of vending machines, turned into where to place them and when to restock. Built with a team.
Land that solves itself
Streaming Wave Function Collapse. Each chunk is solved beside the ones already committed, dropped from memory later, and rebuilt exactly from its seed.