HEI 2.0 — chat client

Frontend for the AI chatbot in «Hva er innafor» 2.0 (HEI 2.0), developed by Helsedirektoratet. It lets Norwegian youth aged 13–19 ask anonymous questions about bodies, feelings, relationships, sexuality and sexual health, and answers them from Helsedirektoratet's own editorial content instead of the open web.

Every answer is grounded in retrieved source material (≈100 articles plus ≈9000 ung.no Q&A pairs), classified by severity on the server, and rewritten in a tone that matches the situation — so a question about something serious gets support and a referral to a help service rather than a cold, factual reply.

This repository contains only the client. The backend lives in llama_chatbot_server_mini and owns retrieval, severity classification, safety routing and answer generation.

---

Repository layout

Path	What it holds
client/	The React app (Create React App + CRACO). See client/README.md.
design/	Design references — redesign mock-up, slide deck, colour palette.
.github/	Azure Static Web Apps CI/CD workflow.

The app is a single UI: client/src/HeiChatClientV3.jsx. Earlier v1/v2 variants and the runtime variant switcher were removed — see git history if you need them back.

Quick start

# 1. Backend, from the llama_chatbot_server_mini repo (separate terminal)
python app.py            # serves on http://localhost:80

# 2. Client
cd client
npm install
npm start                # http://localhost:3000

In development the client talks to http://localhost:80; in a production build it targets the Azure web app. Both are hard-coded in HeiChatClientV3.jsx.

How it fits together

HeiChatClientV3 (React, react-native-web)
   │
   │  GET  /healthz     → is the server done loading indexes?
   │  GET  /documents   → list of source articles ("Kilder" panel)
   │  POST /examples    → topic cards on the landing screen
   │  POST /chat        → the question itself
   │
   ▼  ← Server-Sent Events (answer, refs, related queries, status, systeminfo)
Quart backend (llama_chatbot_server_mini)
   └── LangGraph workflow: retrieve → ground → classify severity → rewrite style

The answer streams back as SSE and is rendered progressively with a typewriter effect. Details of the payload and every event the client understands are in client/README.md.

Deployment

Pushes to main trigger .github/workflows/azure-static-web-apps-ashy-smoke-00a394503.yml, which builds ./client and publishes client/build to Azure Static Web Apps. CI: false is set so build warnings don't fail the job.

History

The repo previously shipped three parallel UIs (HeiChatClient.jsx, HeiChatClientV2.jsx, HeiChatClientV3.jsx) behind a localStorage-backed v1/v2/v3 toggle, plus several HeiChatClient copy*.jsx scratch files. Only V3 is kept. Everything removed is still reachable in git:

git log --diff-filter=D --name-only   # find the commit that deleted a file
git show <commit>^:client/src/HeiChatClientV2.jsx