DeTalks

DeTalks is a mobile-first mental wellness app for students in India. React, Vite and TypeScript on the front, an Express API and PostgreSQL on the back. Live prototype: https://detalks.vercel.app

The tone is calm and non-clinical: warm cream background, forest green, serif headings.

DeTalks is a wellness companion, not a medical or crisis service. In crisis, call Tele-MANAS 14416 (free, 24/7) or 112.

What is in the app

• Email sign-up with an anonymous alias; shared-circle writes require email verification
• Onboarding, consent and a daily pulse check
• Mood tracker, journal and a progress view built from real activity
• Breathing and grounding exercises, plus a practices library
• Talk: text sessions with Disha, an AI companion. Kavach, a safety guardian, screens every message and answers crisis language with verified Indian helplines instead of the model
• Anonymous community circles, also screened by Kavach
• Your data: download everything as JSON, or delete the account for good
• Professional help page with verified helplines
• Installable as a PWA

All screens live in artifacts/detalks/src/pages.

Repository layout

artifacts/detalks        React + Vite + Tailwind web app (wouter routing, framer-motion)
artifacts/api-server     Express 5 API: auth, moods, journal, progress, sessions, circles
artifacts/mockup-sandbox Component preview sandbox
lib/api-spec             OpenAPI spec
lib/api-zod              Generated Zod schemas
lib/api-client-react     Generated API hooks
lib/db                   PostgreSQL schema (Drizzle) and hand-written migrations
scripts                  Workspace scripts

Run it locally

This is a pnpm workspace. Use pnpm, because npm install is blocked by the preinstall script.

pnpm install
pnpm run typecheck

# API (needs a PostgreSQL connection string)
DATABASE_URL=postgres://... PORT=8080 pnpm --filter @workspace/api-server run dev

# Web app (proxies /api to localhost:8080)
pnpm --filter detalks run dev

pnpm run build typechecks and builds every package. pnpm --filter @workspace/api-server test runs the API test suite against an in-memory Postgres (PGlite). The server applies pending migrations on boot.

Environment

• DATABASE_URL — PostgreSQL connection string (required)
• GROQ_API_KEY, GROQ_MODEL - Groq Disha and additive Kavach screening (server-side secrets only)
• APP_ORIGIN, RESEND_API_KEY, VERIFICATION_FROM - email verification, requiring a verified sending domain
• HF_TOKEN — Hugging Face inference token for Disha's replies (optional)
• LLM_API_KEY, LLM_BASE_URL, LLM_MODEL — any OpenAI-compatible provider instead of Hugging Face (optional). Without a key Disha answers with a theme-aware reflective listener
• PORT — listen port (required)
• CORS_ORIGIN — only needed if the frontend is served from another origin

Deployment

The current live app uses Vercel, with vercel.json building the web app and serverless Express API. Set server-side environment variables in the Vercel project; GitHub main is linked to production.

Alternative: render.yaml deploys the whole app as one free Render web service: it builds the frontend, bundles the API, and serves both from a single origin. Provision a free Postgres, set DATABASE_URL (and HF_TOKEN), done.

See RELEASE-CHECKLIST.md for current launch gates. This is a working pilot, not a clinically validated production service.

Contribution split

Contribution plan (50/50): Hassan Rehman focuses on the frontend and UI; Mrunmayee Daware focuses on the remaining software work, including backend/API, integrations, testing, and documentation.

License

MIT