Gabriel Siedler

Cabritos Groceries is a mobile-first collaborative PWA that lets households and groups share grocery lists in real time. Users add items throughout the week and check them off together at the store. The app is built with Next.js 16 (App Router), React 19, and TypeScript on a Neon Postgres back end with Drizzle ORM, and authenticated via Clerk with Google OAuth.

I designed and built the entire stack: the schema (lists, items, members, invites, presence), the REST API, optimistic UI client hooks with setTimeout-based polling, an in-flight mutation guard to prevent race conditions, and a duplicate-prevention layer using normalized names and a partial unique index. Other highlights include invite-link sharing with role-based authorization (owner and guest), a 30-second presence heartbeat with online indicators, autocomplete suggestions, an installable PWA with service worker, rate limiting for public launch, and a polished emerald and slate design system tuned for one-handed in-store use.

Die Das is a mobile-first progressive web app that helps learners master German noun articles (der, die, das) through fast, quiz-style practice sessions. I designed and built the entire product end-to-end, from the visual identity and color system to the offline-capable PWA shell and the audio pipeline.

The app ships with 150 curated nouns, native German pronunciations generated via edge-tts (Conrad Neural voice), accuracy-tiered completion cheers, and a themed learning path with hand-crafted island worlds for each unit. State is persisted locally, so progress survives reloads and offline use. The UI supports English and Portuguese, light and dark themes, and a configurable sound volume. The project demonstrates strong attention to micro-interactions, accessibility, and a polished feel typical of native apps, while remaining a fully installable web app.

Seriesly is a personal TV show tracking web app that helps users follow what they watch, get reminded about upcoming episodes, and discover where to stream each title. Built as a Progressive Web App, it works on mobile and desktop, with installable offline support and native-style push notifications.

I designed and built the project end-to-end. The stack is Next.js 16 with the App Router, React 19, TypeScript, and Tailwind CSS v4. Authentication is handled by Clerk (including Google SSO), data is persisted in PostgreSQL through Drizzle ORM, and series metadata plus regional watch providers are pulled from the TMDB API, with a server-side cache that keeps results available even when upstream calls fail.

A daily Vercel Cron job signs Web Push messages with VAPID and delivers personalized release alerts to each subscriber. The codebase includes unit tests with Vitest and React Testing Library, automated patch versioning via a git hook, and a production deployment on Vercel with environment-driven configuration.

Corre.AI is a sports event discovery platform for Brazil that centralizes running races, triathlons, cycling, swimming, duathlon, and aquathlon competitions in a single monthly calendar. I built it end-to-end as a pnpm monorepo. The front end is a Next.js 15 App Router site with React Server Components, Tailwind CSS 4, and shadcn/ui, featuring a rich filtering experience: users can combine modality, distance ranges per leg (run, bike, swim with independent sliders), state, individual cities, and predefined regions, with URL-synced state, persistence in localStorage, and debounced server actions. The crawler is fully LLM-driven: a Mastra workflow orchestrates Playwright fetches, HTML preprocessing, OpenAI extraction into a strict Drizzle schema, structured validation, and a two-phase deduplication step that combines deterministic matching with semantic LLM comparison to merge the same event across sources with different naming. Mastra also coordinates per-source rate limiting, batch dispatch, retries with exponential backoff, and structured run logs. Storage is PostgreSQL on Neon via Drizzle ORM, with SHA-256 fingerprints for change detection. Deployment runs on Vercel.