AI Engineer World's Fair Schedule Builder

A Sentry-observable Next.js demo app for browsing the AI Engineer World's Fair 2026 schedule and getting AI-assisted recommendations.

Built with Next.js 16, Neon Postgres, pgvector, Vercel AI Gateway, AI SDK 6, and Sentry.

Features

• Schedule grid — Browse AI Engineer World's Fair sessions with day, track, and search filters.
• Talk details — View full session details, speaker info, and save sessions.
• Speaker pages — Browse speakers and their sessions.
• My events view — Save sessions directly from the schedule grid and filter to saved events.
• AI schedule assistant — Right-side chat assistant with semantic schedule search and rendered session cards.
• Tiered AI access — Guests use open models; signed-in users get Claude plus open-model options.
• Sentry observability — Logs, traces, metrics, cache events, DB spans, AI spans, token usage, and quota telemetry.

Tech Stack

• Framework: Next.js 16 App Router, Turbopack, Cache Components
• Database: Neon Postgres + pgvector + Drizzle ORM
• Auth: JWT sessions in cookies
• AI: AI SDK 6 + Vercel AI Gateway
• Embeddings: alibaba/qwen3-embedding-0.6b via Vercel AI Gateway
• UI: shadcn/ui + Tailwind CSS
• Monitoring: Sentry logs, traces, metrics, and AI span streaming

Quick Start

pnpm install

# Configure env vars in .env.local
pnpm db:init
echo "JWT_SECRET=$(openssl rand -base64 32)" >> .env.local
echo "AI_GATEWAY_API_KEY=..." >> .env.local
echo "NEXT_PUBLIC_SENTRY_DSN=https://..." >> .env.local

# Database + data
pnpm db:push
pnpm db:seed
pnpm db:embed

# Dev server
pnpm dev

# Production mode; required for cache behavior
pnpm build && pnpm start

Open http://localhost:3000.

pnpm build prerenders cached pages and requires a valid DATABASE_URL.

Environment Variables

DATABASE_URL=postgres://...        # Neon pooled connection string, sslmode=require
JWT_SECRET=...                     # Any 32+ char string
AI_GATEWAY_API_KEY=...             # Vercel AI Gateway key
NEXT_PUBLIC_SENTRY_DSN=https://... # Sentry DSN

Optional AI usage alert thresholds:

AI_USAGE_ALERT_GUEST_DAILY_REQUESTS=8
AI_USAGE_ALERT_AUTH_DAILY_REQUESTS=75
AI_USAGE_ALERT_DAILY_REQUESTS=100
AI_USAGE_ALERT_DAILY_TOKENS=100000

Scripts

Command	Description
pnpm dev	Start dev server; cache behavior is disabled in dev
pnpm build	Production build
pnpm start	Start production server
pnpm traffic	Generate realistic demo traffic
pnpm db:init	Prompt for Neon/Postgres DATABASE_URL and write .env.local
pnpm db:push	Enable pgvector and apply Drizzle schema
pnpm db:seed	Seed AI Engineer World's Fair 2026 data
pnpm db:embed	Generate pgvector embeddings for schedule search
pnpm db:studio	Open Drizzle Studio
pnpm lint	Run Biome checks
pnpm format	Format with Biome

AI Access Tiers

Tier	Daily requests	Models
Guest	10	GPT OSS, DeepSeek, Qwen, Mistral, Llama
Signed-in users	100	Claude Sonnet/Haiku/Opus plus guest open models
Internal Sentry users	Unlimited	Same as signed-in users

Quota and token usage are persisted in Postgres (ai_usage) and emitted to Sentry as metrics/logs.

Data Fetching

All database access goes through lib/db/queries.ts using plain Drizzle functions.

Pattern	Usage	Caching
"use cache" + cacheTag	Home page, speakers list	Cached with cacheLife
Direct DB query	Talk detail, speaker detail, my schedule	Dynamic
Server Actions	Auth and schedule mutations	Invalidates relevant paths
API Routes	AI chat, AI usage	Dynamic

Observability

Signal	What
page.view metric	Page requests from proxy.ts
cache.miss metric/log	Cached data function executed
account.* logs / account.event metric	Signup, login, logout
schedule.* logs	Saved schedule mutations
ai.chat.requests metric	AI requests by tier/outcome
ai.quota.* metrics	Daily quota usage
ai.tokens.* metrics	Input/output/total model token usage
ai.embedding.* metrics	Embedding requests, tokens, result counts
DB spans	Postgres queries via Sentry postgresJsIntegration
AI spans	AI SDK telemetry via Sentry's default Vercel AI instrumentation and streamGenAiSpans

License

MIT