Add Mindful Coach — Next.js chat UI with coach personas and Vercel deployment

[avaradharaju]

Summary

This PR adds Mindful Coach — a production-style mental wellness coaching app built for the AI Engineer Challenge.

• Next.js 15 chat UI (App Router, Tailwind v4) with light/dark/system theme toggle
• FastAPI backend powered by OpenAI gpt-5 with streaming-ready architecture
• Four coach personas (Professional, Warm Friend, Mindful Guide, Motivational Coach) plus shared safety guardrails
• Creativity slider (Focused → Creative) tuned via system prompt — gpt-5 does not accept custom temperature
• Graceful error handling and a Stop button to cancel in-flight requests
• Vercel services deployment — root vercel.json routes /api/* to FastAPI and everything else to Next.js
• Vibe Check section completed in README (Activities Readme #1–Feature beyond chatgpt #3, reflection questions, and optional improvements)

Links

Live app	https://ai-engineering-challenge-kohl.vercel.app
Health check	https://ai-engineering-challenge-kohl.vercel.app/api/health
Fork	https://github.com/avaradharaju/The-AI-Engineer-Challenge-1
Branch	feat/build-ui

Architecture

Browser → Vercel (Next.js frontend)
              ↓ /api/*
         FastAPI backend (api/index.py)
              ↓
         OpenAI gpt-5

Key files:

• frontend/ — Next.js chat UI, theme, persona/creativity controls
• api/index.py — /api/chat and /api/health endpoints
• api/prompts/ — guardrails, character templates, system prompt builder
• vercel.json — dual-service routing (frontend + backend)

Features

Coach personas

Users pick a coaching style before chatting. All personas share safety guardrails (not a therapist, crisis handling, coaching scope only).

Creativity control

Slider from Focused (~0) to Creative (~100) adjusts tone and style through the system prompt. Vibe check shows clearly different responses at low vs high settings.

UX polish

• Theme toggle (light / dark / system)
• Custom favicon
• User-friendly error banners (no raw 500 messages)
• Cancel button while waiting for a reply

Vibe Check highlights

Documented in the README Vibe Check Your LLM App section:

• Activity Readme #1 — General capability (OOP, summarization, creative writing, math, tone rewrites)
• Activity Readme.md #2 — Real coaching prompts (presentation anxiety, morning routines, burnout) with Motivational Coach at three creativity levels
• Activity Feature beyond chatgpt #3 — Honest handling of capability gaps (no session memory, no live traffic data, no web browsing)
• Optional improvements — Personas, creativity slider, errors, cancel, Vercel fix, theme/favicon

Known limitations

• No cross-session memory (each API call is largely independent)
• No real-time data (traffic, calendars, etc.)
• No external tool / web access for live research
• Coaching guardrails limit scope — not clinical care

Test plan

• Local dev: FastAPI on :8000 + Next.js on :3000
• /api/health returns {"status":"ok"} in production
• Chat works on Vercel with OPENAI_API_KEY set
• Persona + creativity settings persist in localStorage
• Stop button cancels in-flight requests without false error banner
• Theme toggle works across light / dark / system
• Vibe check documented with prompts and responses in README

Deploy notes

Deploy from the repository root (not frontend/). Root Directory in Vercel must be empty so both the Next.js frontend and Python backend deploy via the services block in vercel.json.