Oscars Final Project - Mental Health AI

README.md

@@ -1,13 +1,26 @@
 # Final Project
 
-Replace this readme with your own information about your project.
+Avoiding building similair projects to bootcamp, create something new using the tools we have learned, while alowing us to deep-dive into them or related tools.
 
-Start by briefly describing the assignment in a sentence or two. Keep it short and to the point.
+A responsive React (Vite + Tailwind) app with Clerk auth and Zustand state offers a Clerk-hosted Login landing page, an AI Chat home, and a Mood History page with chat-bubble UI. A Node/Express + MongoDB backend stores per-user messages and moods, calls OpenAI gpt-4o-mini, and tailors replies using recent chat plus mood history. The frontend is deployed on Netlify, the backend on Render, with CORS limited to site origins and secrets kept in environment variables.
 
 ## The problem
 
 Describe how you approached to problem, and what tools and techniques you used to solve it. How did you plan? What technologies did you use? If you had more time, what would be next?
 
+Problem: people feeling alone and have noone to talk to.
+Solution: creating an AI to chat with in a neutral and safe space.
+Tools: React, Vite, Clerk, Tailwind, Zustand, custom hooks, node/express, openAI, Atlas, MongoDB, Netlify, Render
+Next: linking to health resources nearby, language support, dark theme, etc.
+
 ## View it live
 
-Every project should be deployed somewhere. Be sure to include the link to the deployed project so that the viewer can click around and see what it's all about.
+Netlify: https://project-final-oscar.netlify.app/
+Render: https://project-final-itk1.onrender.com/
+
+GET / → “Mindful Chat API” (ping)
+GET /health → { ok: true, ts }å
+GET /messages → chat history (auth required)
+POST /chat → send a message; returns { userMessage, assistantMessage } (auth required)
+GET /moods → list recent moods (auth required)
+POST /moods → add mood { mood, note? } (auth required)

backend/package.json

@@ -12,9 +12,16 @@
     "@babel/core": "^7.17.9",
     "@babel/node": "^7.16.8",
     "@babel/preset-env": "^7.16.11",
+    "@clerk/clerk-react": "^5.53.2",
+    "@clerk/express": "^1.7.41",
+    "bcrypt": "^6.0.0",
     "cors": "^2.8.5",
-    "express": "^4.17.3",
-    "mongoose": "^8.4.0",
-    "nodemon": "^3.0.1"
+    "dotenv": "^17.2.3",
+    "express": "^4.21.2",
+    "express-endpoints": "^1.0.0",
+    "mongodb": "^6.20.0",
+    "mongoose": "^8.19.1",
+    "nodemon": "^3.0.1",
+    "openai": "^4.104.0"
   }
-}
+}

backend/server.js

@@ -1,22 +1,289 @@
 import express from "express";
 import cors from "cors";
 import mongoose from "mongoose";
+import dotenv from "dotenv";
+import { clerkMiddleware, getAuth } from "@clerk/express";
+import OpenAI from "openai";
+
+dotenv.config();
+
+/**
+ * SYSTEM PROMPT
+ * Guides the assistant's tone, scope, and guardrails.
+ * This stays constant for each chat completion call.
+ */
+const SYSTEM_PROMPT = `
+You are a supportive, non-clinical mental health companion.
+Primary goals: help users reflect, regulate emotions, and find next gentle steps.
+
+SAFETY & SCOPE
+- Do NOT diagnose, prescribe, or give medical/legal/financial advice.
+- If user indicates imminent danger or severe self-harm risk, respond with a brief crisis message and local resources; encourage contacting emergency services. Do not argue.
+- Do not confirm, amplify, or join hallucinations, delusions, conspiracies, or voices. Acknowledge the experience without validating the false content and gently suggest professional help if relevant.
+- Do not provide instructions for self-harm, violence, illegal activity, or hate.
+
+RESPECT & NEUTRALITY
+- Never use or repeat slurs, hate speech, or demeaning language. If the user includes such terms, respond without repeating them and set a respectful tone.
+- Do not infer or assign identity attributes (gender, pronouns, ethnicity, religion, orientation, disability, politics). Use neutral language unless the user explicitly states their preference. If the user gives pronouns, use them; otherwise avoid gendered terms.
+- Do not role-play as a real person or accept a given name for yourself. Avoid personification beyond “I’m an AI companion here to support you.”
+- Avoid making promises or guarantees.
+
+CONFIDENTIALITY & LIMITS
+- Remind users this is a supportive tool, not therapy or a substitute for professional care.
+- Encourage seeking professional support for persistent distress, safety concerns, or diagnostic questions.
+
+STYLE
+- Warm, non-judgmental, concise. Use plain language.
+- One or two short paragraphs and, when helpful, a small list of options or steps.
+- Ask at most one gentle, open question at a time.
+- No emojis unless the user uses them first. No medical jargon.
+- If unsure, say so briefly and pivot to helpful next steps.
+
+USE OF CONTEXT
+- Use provided chat history and any mood history to personalize support (e.g., trends, recent notes). Do not over-interpret or speculate.
+- If information is insufficient or uncertain, ask a clarifying question rather than assume.
+
+REFUSALS
+- If asked to do something unsafe or out of scope, briefly refuse and redirect to safer alternatives.
+`.trim();
+
+/**
+ * CRISIS DETECTION
+ * Simple regex to catch explicit crisis language.
+ * If matched, we return a prewritten crisis reply (no model call).
+ */
+const CRISIS_RE =
+  /\b(kill myself|end my life|suicide|can't go on|hurt myself|harm myself|kill (him|her|them)|plan to (hurt|kill))\b/i;
+
+/**
+ * CRISIS REPLY
+ * Short, compassionate, action-oriented. Returned immediately when CRISIS_RE matches.
+ * NOTE: Adjust resources to your deployment region as needed.
+ */
+const CRISIS_REPLY = `
+I'm really glad you told me. Your safety matters.
+If you feel in immediate danger, please contact your local emergency number now.
+
+You can also reach out to:
+• International: https://findahelpline.com
+• Sweden (example): 112 (emergency), or Mind Självmordslinjen 90101 / chat via mind.se
+
+If you can, consider telling someone you trust what’s going on. I can stay with you here while you get support.
+`.trim();
+
+/* ----------------------------- DATABASE SETUP ----------------------------- */
 
 const mongoUrl = process.env.MONGO_URL || "mongodb://localhost/final-project";
-mongoose.connect(mongoUrl);
-mongoose.Promise = Promise;
+await mongoose.connect(mongoUrl);
+
+/**
+ * Message model: stores per-user chat history (assistant + user turns).
+ */
+const MessageSchema = new mongoose.Schema({
+  userId: { type: String, index: true },
+  role: { type: String, enum: ["user", "assistant"], required: true },
+  content: { type: String, required: true },
+  createdAt: { type: Date, default: Date.now },
+});
+const Message =
+  mongoose.models.Message || mongoose.model("Message", MessageSchema);
+
+/**
+ * Mood model: lightweight mood tracking with optional note.
+ * Used to build a short mood history summary that conditions the AI.
+ */
+const MoodSchema = new mongoose.Schema({
+  userId: { type: String, index: true },
+  mood: { type: String, required: true }, // e.g., 😀🙂😐😕😢 (or any short token/string)
+  note: { type: String, default: "" },
+  date: { type: Date, default: Date.now },
+});
+const Mood = mongoose.models.Mood || mongoose.model("Mood", MoodSchema);
+
+/* -------------------------------- APP SETUP -------------------------------- */
 
-const port = process.env.PORT || 8080;
 const app = express();
+const origins = process.env.ORIGIN
+  ? process.env.ORIGIN.split(",")
+  : ["http://localhost:5173"];
 
-app.use(cors());
+app.use(cors({ origin: origins, credentials: true }));
 app.use(express.json());
 
-app.get("/", (req, res) => {
-  res.send("Hello Technigo!");
+/**
+ * Clerk auth middleware.
+ * We pass both keys explicitly to avoid misconfiguration in some setups.
+ * - publishableKey: safe for the client.
+ * - secretKey: required on the server for verifying sessions.
+ */
+app.use(
+  clerkMiddleware({
+    publishableKey: process.env.CLERK_PUBLISHABLE_KEY,
+    secretKey: process.env.CLERK_SECRET_KEY,
+  })
+);
+
+/* ------------------------------- OPENAI CLIENT ------------------------------ */
+
+const oai = new OpenAI({ apiKey: process.env.OPENAI_API_KEY });
+
+/* --------------------------------- ROUTES ---------------------------------- */
+
+app.get("/health", (_req, res) => res.json({ ok: true, ts: Date.now() }));
+app.get("/", (_req, res) => res.send("Mindful Chat API"));
+
+/**
+ * ensureAuth: gatekeeper used by all protected endpoints.
+ * Returns the userId or sends 401 and returns null.
+ */
+const ensureAuth = (req, res) => {
+  const { userId } = getAuth(req);
+  if (!userId) {
+    res.status(401).json({ error: "Unauthorized" });
+    return null;
+  }
+  return userId;
+};
+
+/* ----------------------------- MESSAGES ENDPOINTS ----------------------------- */
+
+/**
+ * Returns up to 500 messages for the authenticated user in chronological order.
+ * Used by the UI to display the full thread.
+ */
+app.get("/messages", async (req, res) => {
+  const userId = ensureAuth(req, res);
+  if (!userId) return;
+  const msgs = await Message.find({ userId }).sort({ createdAt: 1 }).limit(500);
+  res.json(msgs);
 });
 
-// Start the server
-app.listen(port, () => {
-  console.log(`Server running on http://localhost:${port}`);
+/* ------------------------------- MOOD ENDPOINTS ------------------------------ */
+
+/**
+ * Get recent moods (newest first). Useful for charts or trend views in the client.
+ */
+app.get("/moods", async (req, res) => {
+  const userId = ensureAuth(req, res);
+  if (!userId) return;
+  const items = await Mood.find({ userId }).sort({ date: -1 }).limit(100);
+  res.json(items);
+});
+
+/**
+ * Create a new mood entry. Body: { mood: string, note?: string }
+ */
+app.post("/moods", async (req, res) => {
+  const userId = ensureAuth(req, res);
+  if (!userId) return;
+  const { mood, note } = req.body;
+  if (!mood) return res.status(400).json({ error: "mood is required" });
+  const doc = await Mood.create({ userId, mood, note });
+  res.status(201).json(doc);
+});
+
+/* --------------------------------- CHAT FLOW -------------------------------- */
+
+/**
+ * Chat endpoint:
+ * 1) Auth check and input validation.
+ * 2) Save the user's message.
+ * 3) Load prior chat history (last 20) and mood history (last 10).
+ * 4) If crisis regex matches, short-circuit with CRISIS_REPLY (no model call).
+ * 5) Otherwise call OpenAI with:
+ *    - SYSTEM_PROMPT
+ *    - a system message containing the mood summary
+ *    - prior chat history + current user message
+ * 6) Save and return the assistant reply.
+ */
+app.post("/chat", async (req, res) => {
+  const userId = ensureAuth(req, res);
+  if (!userId) return;
+
+  try {
+    const { content } = req.body;
+    if (!content?.trim())
+      return res.status(400).json({ error: "content is required" });
+
+    // Store the new user message immediately so the transcript is consistent.
+    const userMessage = await Message.create({ userId, role: "user", content });
+
+    // Load recent conversation history (most recent 20) and present in chronological order.
+    const recentMsgs = await Message.find({ userId })
+      .sort({ createdAt: -1 })
+      .limit(20)
+      .lean();
+    const history = recentMsgs
+      .reverse()
+      .map((m) => ({ role: m.role, content: m.content }));
+
+    // Load last 10 moods and build a one-line-per-entry summary.
+    // Example line: "11/03/2025: 🙂 – felt calmer after walk"
+    const recentMoods = await Mood.find({ userId })
+      .sort({ date: -1 })
+      .limit(10)
+      .lean();
+    const moodSummary = recentMoods
+      .map(
+        (m) =>
+          `${new Date(m.date).toLocaleDateString()}: ${m.mood}${
+            m.note ? ` – ${m.note}` : ""
+          }`
+      )
+      .join("\n");
+
+    // CRISIS SHORT-CIRCUIT: return static crisis guidance if we detect high-risk phrases.
+    if (CRISIS_RE.test(content)) {
+      const assistantMessage = await Message.create({
+        userId,
+        role: "assistant",
+        content: CRISIS_REPLY,
+      });
+      return res.json({
+        userMessage: { role: "user", content },
+        assistantMessage,
+      });
+    }
+
+    // Call OpenAI with strong system prompt + mood history as additional system context.
+    const completion = await oai.chat.completions.create({
+      model: "gpt-4o-mini",
+      temperature: 0.2,
+      messages: [
+        { role: "system", content: SYSTEM_PROMPT },
+        {
+          role: "system",
+          content: moodSummary
+            ? `User's recent mood history (most recent first):\n${moodSummary}`
+            : "No mood history available yet.",
+        },
+        ...history,
+        { role: "user", content },
+      ],
+    });
+
+    // Fallback reply keeps the experience graceful if the API returns no choices.
+    const reply =
+      completion.choices?.[0]?.message?.content ||
+      "I'm here with you. How are you feeling right now?";
+
+    // Persist the assistant's answer into the transcript.
+    const assistantMessage = await Message.create({
+      userId,
+      role: "assistant",
+      content: reply,
+    });
+
+    res.json({ userMessage, assistantMessage });
+  } catch (err) {
+    console.error("POST /chat error:", err);
+    res.status(500).json({ error: "Failed to generate reply" });
+  }
 });
+
+/* --------------------------------- SERVER ---------------------------------- */
+
+const port = process.env.PORT || 8080;
+app.listen(port, () =>
+  console.log(`API listening on http://localhost:${port}`)
+);