ADHD Learning Companion
A structured, cyclical AI learning companion: explanation → feedback → interest-based
personalization → technical question → answer evaluation → next-stage expansion, repeating.
Originally a Streamlit prototype, rebuilt as a Next.js app deployable on Vercel.
Live: https://adhd-learning-companion.vercel.app
Why this is a bounded conversational agent, not a chatbot
The model never drives the conversation. A fixed client-side state machine (app/page.tsx)
decides which of five stages is active — question → feedback → interest → questionPhase → answer → (back to feedback) — and only calls the model with a stage-specific system prompt
for that one turn. The LLM produces content within a stage; it never chooses the next stage.
Stack
- Next.js 14 (App Router, TypeScript)
- Tailwind CSS
- OpenRouter chat completions API (
google/gemini-2.5-flash-lite), called server-side only
from app/api/chat/route.ts — the API key never reaches the browser.
Local development
npm install
cp .env.local.example .env.local # fill in OPENROUTER_API_KEY
npm run dev
Open http://localhost:3000.
Deployment (Vercel)
- Push this repo to GitHub.
- Import it at https://vercel.com/new.
- Add environment variable
OPENROUTER_API_KEY in Project Settings → Environment Variables.
- Deploy.
Security note
OPENROUTER_API_KEY is read only inside app/api/chat/route.ts (a server-side route
handler) and is never sent to the client. .env.local is gitignored.
Disclaimer
Educational tool only. Not a clinical diagnostic system, psychological assessment tool, or
medical intervention platform.
