Soba: Voice AI Emotional Companion

Over 166 million young people live with a mental health condition, yet most go untreated. Many want to talk but fear judgment, feel like a burden, or don't know where to start. By the time anyone notices, help is already late. Soba is a voice-first AI emotional companion built around three steps: Listen, Support, Connect. Users simply talk out loud, and Soba answers in real time with a warm voice that adapts to their mood. It guides breathing and grounding when things feel overwhelming. When a conversation shows signs that more help is needed, Soba offers to connect the user with a trusted person, guardian, or professional support, always with consent. Every reply passes through a three-stage safety pipeline (assess, reply, check) that fails closed. Conversations are sorted into green, yellow, and red risk tiers, so Soba knows when to keep supporting, when to gently suggest human help, and when to surface emergency resources. The working MVP is Soba Web. Users can talk to Soba, track mood patterns, keep journal reflections, personalize Soba's personality and voice, and choose which memories it keeps. A separate guardian view shows only the wellbeing trends and safety signals the user has approved, never private conversations. Privacy is built in: consent first, data minimization, and no raw transcripts by default. We are also designing the Soba Companion, a plush bear with a built-in voice module for a more present, physical experience. The hardware is currently in the planning and validation phase. Soba is a companion, not a therapist, and it does not diagnose. Its role is to make the first conversation easier and bring the right human support closer.