Skip to Main Content

Vertalt

Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Vertalt — Demo video

**SELA — Your Story. Your Meaning. Your Next Step.** SELA is a privacy-first, bilingual AI wellbeing companion built around a simple problem: people often struggle to express what they are experiencing, understand their feelings over time, and prepare their stories when they decide to seek support. Traditional AI conversations can introduce another problem: an interpretation may subtly change what someone meant. "Sometimes I need to be alone" can become "I always avoid people." If this misunderstanding enters persistent memory, summaries, or shared information, it may follow the person beyond the original conversation. **SELA introduces the Meaning Integrity Engine (MIE)** to address this challenge. MIE connects original, versioned statements to AI interpretations, memories, consultation summaries, and shared information. Users can review, accept, correct, or reject interpretations. When underlying meaning changes, SELA tracks affected dependencies, marks outdated information for review, and prevents invalidated content from being treated as current. Beyond conversational AI, SELA includes: • Bilingual AI-powered listening and personal reflection. • Private journaling, check-ins, and a source-linked Life Context Timeline. • Personal Action Bridge for optional, self-chosen next steps. • Care Passport, a user-approved consultation preparation sheet. • Consent-based sharing and simulated professional care requests. • Privacy controls, encrypted personal records, and safety guardrails. **Built with Meta Muse Spark and Muse Wonder through Meta Model API**, SELA combines generative AI with source verification and deterministic safeguards implemented in Go. Its frontend uses React and TypeScript, backed by PostgreSQL and a versioned dependency graph. Our live Meta evaluation includes 128 synthetic cases. The evaluation demonstrated that the verification pipeline could reject deliberately distorted interpretations, while also revealing false-positive tradeoffs and the limitations of our small synthetic dataset. SELA is designed to reduce the burden of repeatedly explaining personal experiences while keeping users—not AI—in control of their meaning, information, and next steps. The demo uses synthetic data and fictional professional partners. SELA does not provide diagnosis, therapy, emergency response, or clinically validated treatment. **Our vision: AI should help people express their stories, not rewrite them.**

Team