Yalla
A B2B ticketing and event platform running in production in Brazil, with tickets issued on-chain and a transfer policy the producer cannot rewrite after the fact.
- Role
- Full stack, solo — everything but mobile
- Year
- 2026 · Ongoing
- Stack
- Rust, Soroban, Stellar, TypeScript, Next.js, Express, Prisma, PostgreSQL, Redis, Privy, Stripe Connect, Mercado Pago, AWS ECS, Prometheus, Turborepo
Producers create events, sell tickets in lots, run promo codes and food-and-beverage vouchers, and check people in at the door — the whole lifecycle without leaving the platform. Tickets are issued on Stellar and the transfer policy is written on-chain and immutable, so the rules a buyer agreed to outlive any change of mind on the platform side.
I am the only person building it, and the scope is the whole thing: the Soroban contracts in Rust, the Express and Prisma API, the Next.js producer panel, the brand and design system it is drawn in, the AWS pipeline it deploys through, and the operational reports the partners read after an event. 608 commits since March 2026. The only piece I did not write is the mobile app.
- Wrote both Soroban contracts in Rust — ticket and voucher — upgradeable, with a minter allowlist and the admin and upgrade authority kept deliberately disjoint, so the key that mints can never be the key that replaces the code.
- Ran the platform's first event end to end in production — Prelúdio, 16 May 2026: 113 tickets across 8 lots, 98 people through the door, and a peak of 39 check-ins inside a single hour. Wrote the exports and the report the partners read afterwards.
- Built the voucher economy behind the bar: 269 vouchers issued and 257 redeemed on the night — 95.5% redemption, zero expired.
- Wired both payment rails the market needs: Mercado Pago, where PIX carried 92.5% of pre-event sales, and Stripe Connect for producer payouts.
- Own the delivery pipeline end to end: five GitHub Actions workflows (CI, e2e, security, staging, production), separate ECS task definitions for API and worker in each environment, Prometheus alerting, and ten runbooks including rollback and the cutover for upgrading a live contract.
- Own reliability in production: an inventory reconciler for orphaned holds, expiry that gives back what a pending order reserved, and a Stellar write retry narrowed to `txBadSeq` alone — so a real failure is never retried blindly.