Live Show Radar

An AI agent that watches the artists you follow and tells you the moment any of them announces a show in India — Bengaluru first — with the venue, date, early-bird status and deadline, so you can buy the same day.

Live URL: https://live-show-radar-sigma.vercel.app · Repo: https://github.com/pratikshak-maker/Live-show-radar

Screenshots: placeholder — Radar (dark/light), show drawer, Activity live log, map view.

The 5-step workflow

1. Artists — ships with a 40-artist starter list; add by name (MusicBrainz autocomplete), import from Last.fm, or connect Spotify.
2. Agent scans — an external scheduler hits the app every 30 minutes; each run checks a small batch, or press Check now and watch it work live.
3. Alert — one Gmail email per run (Bengaluru + early-bird items first) plus an in-app bell.
4. Decide — open a show for the AI "Should I go?" brief and tier table with live IST countdowns.
5. Book it — I'm going creates Google Calendar events (concert, early-bird deadline, on-sale); without Google you get pre-filled calendar links.

No signup, no login wall, no onboarding. Every integration is an optional upgrade.

Components Map

Full diagrams and the component table are in docs/COMPONENTS_MAP.md.

flowchart LR
  Sched[cron-job.org / Vercel Cron] --> Cron["/api/cron/scan"] --> Agent
  UI[Next.js UI] --> API[API routes] --> Agent
  Agent[Agent: discover → enrich → extract → validate → reconcile → notify]
  Agent --> Gemini & TM[Ticketmaster] & SB[ScrapingBee]
  Agent --> DB[(Neon Postgres)]
  Agent --> Gmail[Gmail API / SMTP]
  API --> Cal[Google Calendar]

Tech stack

Next.js 16 (App Router) · React 19 · TypeScript (strict) · Tailwind CSS 4 · Radix UI + lucide-react · Neon Postgres (@neondatabase/serverless, server-only, parameterised SQL) · @google/genai (Gemini Flash only) · googleapis · nodemailer · zod · date-fns(-tz) · cheerio · fastest-levenshtein · Leaflet + OpenStreetMap · Vitest · ESLint + Prettier · Vercel.

External APIs (10)

API	What it does here
Gemini (Google Search grounding + JSON mode)	Finds announcements, extracts structured shows, writes the "Should I go?" brief, answers questions
ScrapingBee (HTML + Google Search)	Reads live ticketing pages for tiers/deadlines; fallback search
Gmail API	Sends alert emails as you (SMTP app-password fallback)
Google Calendar API	Creates/deletes concert, early-bird and reminder events
Ticketmaster Discovery	Official listings as a second source
Spotify Web API	Imports followed artists (allow-listed accounts only)
Last.fm	Imports top artists from a username
MusicBrainz	Artist autocomplete + disambiguation
Deezer	Artist photos (≥0.85 name match)
OpenStreetMap (Nominatim + tiles)	Venue geocoding and the map

How scanning works

Per artist: A discover (Gemini grounded search, Ticketmaster, ScrapingBee Google fallback) → B enrich (scrape allow-listed ticketing pages under a daily credit budget; JSON-LD parsed) → C extract (Gemini JSON mode, zod-validated, one repair retry) → D validate (deterministic: right artist, in India, sane date, every cited URL must be one the model was given) with a 0–1 confidence score → E reconcile (dedupe incl. ±1 day, merge tiers/sources, derive early-bird status, detect changes, priority score) → F notify (idempotent in-app + one email per run). Runs are single-flight (Postgres lock), time-budgeted, and every step writes a plain-English line to the live Activity log.

Run locally

npm install
cp .env.example .env.local    # everything is optional
npm run dev                   # http://localhost:3000

With no keys you'll see the Setup screen; add DATABASE_URL (Neon) first (SETUP.md). Other commands:

npm run lint && npm run typecheck && npm test && npm run build   # all pass with NO env vars
npm run check:integrations    # ✅/❌ + latency for each configured API
npm run seed                  # load the starter artists
npm run check:secrets         # fails if a secret is exposed via NEXT_PUBLIC_

Evaluation criteria → where it shows up

Criterion	How it's answered
UI/UX quality	"Stage at night" design system, ticket-stub cards with live countdowns, radar sweep, map, skeletons, empty/error states, mobile-first
Complexity	6-stage multi-source agent, grounded search + scraping + structured extraction, deterministic validation, confidence + priority scoring, change detection, idempotent alerts, live log
User experience	Value on first screen; one-tap Book / I'm going / Remind me; Undo instead of confirm dialogs
External API usage	10 real integrations, each in lib/integrations/<name>.ts with typed Results
Nothing mandatory	No signup/login/onboarding; every integration optional with a graceful "Not connected yet" state

Known limitations

• Early-bird data is only as good as what ticketing pages publish; many pages hide tiers behind JavaScript or lists changes late.
• Gemini grounded search can miss announcements that only exist on Instagram; the validator prefers dropping to guessing.
• Spotify connect only works for accounts the app owner has allow-listed (Development Mode), and needs a Premium owner.
• Free tiers apply (Gemini daily quota, ScrapingBee credits, Neon free tier, Vercel Hobby limits); the app meters and degrades instead of failing.
• Ticketmaster has thin India coverage; an empty result is normal.
• Scan latency is bounded by the scheduler (30 min) and Vercel's function time limit (batches of 4 artists per run).
• Database code is tested end to end against PGlite (Postgres in WASM) and smoke-tested against the live Neon database.