Sumi

Inspiration

We were inspired by the stories of young people who face homelessness and are forced to retell their trauma repeatedly just to access basic services. As founders coming from backgrounds in mental health, AI, and social impact, we wanted to create a tool that listens once, and helps without judgment or pressure. Sumi was born from the belief that care should start with dignity, not data collection.

What it does

Sumi is a trauma-informed AI care navigator that helps youth experiencing housing instability find housing, food, healthcare, and community resources safely and on their own terms. It builds trust first - showing local, low-barrier resources like art therapy or LGBTQ+ centers without requiring intake forms. When the user is ready, Sumi gently guides them through a single conversational (not interrogational) intake and creates a secure, youth-controlled digital profile that can auto-fill agency forms and connect them directly to verified services.

How we built it

We built Sumi as a web-based prototype using React, Supabase, and Lovable AI (OpenAI/Gemini API). The system runs locally first for privacy, encrypting all user data and only sharing it with consent. The conversational flow was designed using a psychological blueprint rooted in trauma-informed care, motivational interviewing, and safety planning. Voice interaction (via ElevenLabs) and sentiment-based tone adjustment make Sumi feel like a caring guide rather than a chatbot.

Challenges we ran into

• Not being able to integrate to HMIS (weekend hours).
• Designing trauma-informed AI prompts that balance empathy with precision.
• Building a privacy-first architecture that stores data locally but still supports referrals.
• Integrating multiple resource databases and ensuring accurate, inclusive filtering (LGBTQ+, BIPOC, accessibility).
• Testing conversational flows that felt genuinely safe for users in distress.

Accomplishments that we're proud of

• Developed a working prototype that demonstrates real-time conversation, local-first storage, and smart referrals.
• Created a trauma-informed, evidence based, interaction framework grounded in psychological science.
• Designed a cross-disciplinary product combining AI, psychology, and UX for social impact.
• Partnered with Blisstherapy.org to offer a free therapy session for users needing mental health support.

What we learned

• AI can be compassionate, if designed with human psychology in mind.
• Trust is the first barrier in social care, not technology.
• Collaboration between technical and mental health teams creates more ethical, effective products.
• Simplicity (like letting users say “not now”) can be more powerful than complex algorithms.

What's next for Sumi

• Pilot Sumi with the California Homeless Youth Project (CHYP) and youth shelters in the Bay Area.
• Integrate with official HMIS databases for secure referrals.
• Expand to multilingual versions.
• Develop an open API for nonprofits and local governments.
• Scale nationally as a privacy-first AI companion that bridges care gaps and reduces retraumatization.

Built With

• css
• elevenlabs
• gemini
• lovable
• natural-language-processing
• openai
• postgresql
• react
• shadcn/ui
• tailwind
• typescript
• vite