Kintsu: Decolonizing Mental Health with Cultural AI

Elevator Pitch

Kintsu is a privacy-first cultural translator that helps second-generation immigrants articulate mental health struggles to their parents using heritage proverbs instead of clinical jargon. By reframing concepts like "burnout" into values like "duty" or "restoration" via AI, we turn potential family conflicts into moments of safe connection.

Problem Statement

For diaspora youth, mental health terms like "anxiety" or "boundaries" often translate to "laziness" or "disrespect" in their parents' native language. This linguistic divide forces a choice between suffering in silence or risking family alienation, leaving millions without a safe way to ask for help. Existing AI tools often fail these users by relying on Western clinical data that ignores non-Western family dynamics.

Our Solution

We built a web application that acts as an ego-protecting diplomat. Users anonymously input their raw emotion, and our RAG-powered engine reinterprets it using authenticated cultural proverbs and folklore. The tool generates a digital "Bridge Card" containing a culturally validated script that users can send directly via WhatsApp (our main focus, but other chat apps may also be used), effectively translating their mental health needs into wisdom their parents already respect. It includes safety guardrails for crisis detection and a "Data Philanthropy" loop that anonymizes inputs to help researchers reduce bias in future AI models.

How It Works (Technical Overview)

The application is built on Next.js and deployed on Vercel. We use the Gemini API in strict JSON mode to parse user sentiment against a curated dataset of cultural proverbs (RAG architecture), ensuring citations are authentic to prevent hallucinations. We utilize Supabase for ephemeral data storage, ensuring 24-hour auto-deletion for privacy and using Row Level Security for encryption. The frontend uses Tailwind CSS to create an accessible "digital ceramics" interface that avoids clinical sterility. We also integrated Open Graph dynamic image generation to ensure shared links look welcoming in messaging apps.

What Makes It Unique / Innovative

Unlike generic therapy bots that use Western clinical models, Kintsu specifically leverages non-Western wisdom to validate emotions, using heritage as a bridge rather than a barrier. Our "Data Philanthropy" engine is also unique: every interaction contributes to an anonymized aggregate dataset that helps researchers decolonize future AI models, allowing users to feel they are contributing to a global solution just by surviving their personal struggles.

Challenges Faced

Ensuring the AI did not hallucinate fake proverbs was critical for maintaining trust with older generations. We overcame this by implementing a strict JSON schema that forces the LLM to cite specific sources (like religious texts or folklore) and by implementing a confidence score filter. Balancing data utility with user privacy was also difficult, which we solved by making all link data ephemeral and self-destructing after 24 hours.

Future Scope

We plan to launch a "Clinician Dashboard" where therapists can generate these scripts during sessions to aid family therapy. We also aim to expand our folklore dataset to include Indigenous and Eastern European cultures and implement voice-to-text features to support oral storytelling traditions.

Safety & Impact

Safety is paramount. We implemented client-side and server-side logic to detect self-harm or crisis keywords immediately. If detected, the system overrides the AI generation to display local crisis resources and hotlines instead of a proverb. Furthermore, the application includes a disclaimer clarifying it is a cultural tool, not a medical device.

Built With

• gemini
• next.js
• postgresql
• react
• supabase
• typescript