JARVIS AI Doomsday

The problem

Voice assistants feel like a search box that talks. They are faceless, tied to one company’s AI, and quietly collect far more than they need.

The approach

Give the assistant a presence. JARVIS AI Doomsday puts a cinematic, fully rendered 3D character at the centre of the screen and lets it act as the interface — listening, thinking and answering in character. Behind it sits a provider-agnostic AI layer, so users bring the model they trust and pay only for what they use. Privacy is a feature, not a footnote: speech, vision and memory run on the phone wherever possible, and anything that leaves it needs the user’s say-so.

What I built

• A native Kotlin + Jetpack Compose app with a Filament render stage for the animated glTF character
• Wake word, clap and shake triggers, on-device speech recognition and spoken replies
• A pluggable provider layer for seven AI services, with keys sealed by the Android hardware keystore
• On-device vision with CameraX and ML Kit object detection, labelling and text recognition
• Explicit-only memory with a relationship graph, reminders, a diary, message drafting and confirmed calls
• Guardian safety tools: an emergency contact, location sharing, quiet hours and an outbound-activity log
• A Ktor companion backend for streaming chat, live data and natural text-to-speech