Zayyan OS

What it is

Named after my son, who is also its first user. Applications define a robot experience. Zayyan OS provides the reusable capabilities. Hardware adapters supply physical I/O. No core service may depend on a specific robot, board, sensor, or cloud. The first application is a teddy that sees a person through a USB camera, wakes, greets in Bangla, holds a spoken conversation, and returns to idle.

How it works

• A proactive perception loop: camera → face, gaze, emotion, object, and activity detection with lightweight OpenCV models → a world model → a deterministic behaviour engine that decides whether and when to speak, with cooldowns, quiet hours, and an interruption budget. The LLM proposes the words; code makes the decision.
• Mediated tools: the model can render rich HTML, SVG, and canvas on the robot's screen or request a photo, but every tool call is executed by deterministic code behind a safety gate. Vision is off by default because it sends a picture to the cloud.
• Two personas: a child companion and a student tutor, with automatic switching that defaults to the child-safe persona on any uncertainty.
• Tools exposed over the Model Context Protocol, and external MCP servers consumed, starting with a keyless Bengali-first Wikipedia lookup.
• A browser-based live demo drives the real product code with a web UI instead of the teddy.

Status

Planning and architecture baseline plus the first runnable slice, in simulation and on Jetson. Documentation-first: no subsystem is implemented before its specification and acceptance criteria exist.