The problem
“I'm about to place this bet. Am I missing something important?”
Case study · Web app · Mobile-first
ParlayCheck
Check your parlay before you bet.
Upload screenshots of a parlay you've already built, and AI looks up current, sourced information for every selection (injuries, lineups, form and news) without making predictions.
- Type
- Web app
- Status
- Live
- Year
- 2026
parlaycheck.io
Who it's for
Sports bettors in Mexico and the US who already built a parlay and want a second look right before betting.
Informs, never predicts
ParlayCheck is deliberately not a tipster, a pick generator or an odds calculator. It shows what recent information matters, which selection it affects and where it comes from. The user decides.
What it does
Features
AI screenshot reading
Up to 6 screenshots and 15 selections, merging overlapping captures. The user confirms or fixes the selections, and instructions hidden in images are ignored.
Research with verified sources
Gemini with Google Search grounding, with OpenAI web search as a fallback. Topics adapt to each market, and every finding gets one of four statuses.
Check packs with Stripe
Packs of Checks through Stripe Checkout. The currency depends on the country, not the language, and delivery is idempotent.
Passwordless account
Magic-link sign-in, a history of every Check, pack progress and account deletion with a 30-day grace period.
Bilingual guides
Nine guides in English and Spanish with structured data, generated social images and an hreflang sitemap.
The journey
How it works
- 01
Upload screenshots of the parlay
- 02
Confirm or fix the selections
- 03
Check out with Stripe
- 04
AI researches every selection
- 05
Read the findings with their sources
Under the hood
Engineering highlights
Built against hallucinations
Source URLs are validated against the search tool's real results, not the model's text. Findings without evidence or with predictive language are discarded.
Provider fallback
If one AI provider fails, the next one takes over. Errors are classified, and every attempt logs latency, tokens and estimated cost.
Benchmarked providers
A Gemini vs OpenAI benchmark, including prompt-injection cases, guides which model does what.
Fail-safe pricing
Price is chosen by country, separately from the language, so a user is never shown one currency and charged another.
Frontend
- Next.js 16
- React 19
- TypeScript
- Tailwind CSS 4
- shadcn/ui
Backend
- Server Actions
- Route handlers
- Zod at every trust boundary
Data
- Prisma 7
- Supabase Postgres (RLS)
- Private Storage
AI & services
- Gemini
- OpenAI
- Stripe
- Supabase Auth
- Resend
- Vercel