Applied
Engineering
Results
samuel-torres.com
Role
Backend development, WordPress plugin architecture and an AI verification pipeline for a live, nationwide sticker-exchange event.
Core stack
WordPress · 2 custom PHP plugins · MySQL · Tesseract OCR · Groq API (LLaMA 3.3 70B) · Cron · AlmaLinux VPS
Context
Panini ran an in-store sticker-exchange event across Mexico, Panini Fan Zone FWC 2026, including a push toward a Guinness World Records attempt for the largest sticker exchange. The event needed a folio-and-prize system that could issue and track a fixed pool of 80,000 folios, verify submitted purchase receipts without a person reading every one by hand, and register participants for the record attempt without letting the same person get counted twice under different activities.
What I did
Built a custom WordPress plugin (panini-registro) issuing and tracking a fixed pool of 80,000 folios with QR code generation for each registration, running live on an AlmaLinux VPS.
Instead of manual review, submitted receipts go through Tesseract OCR first, then a Groq-hosted LLaMA 3.3 70B model reads and validates the extracted text against the promotion's rules. The pipeline runs unattended via cron every 2 minutes, live in production.
The record-attempt registration needed its own uniqueness rule — one person, one entry, scoped to that specific activity — so it got its own database schema rather than reusing the folio plugin's checks. The first version's uniqueness index checked email globally instead of per-activity, which let the same email register twice under different activities. I migrated the schema to a composite unique index scoped to the activity, added a guard that checks the existing schema before altering it, and bumped the plugin from v1.1.0 to v1.1.1.
Wrote a standalone verifier script and a 15-check test suite, including a concurrency test with 50 simultaneous workers hitting the same registration path, to confirm exactly one registration succeeds per slot with zero server errors before treating any of it as done.
Result
The folio and QR system ran live in production managing the full 80,000-folio pool, with the OCR → LLM pipeline processing submissions unattended. The Guinness-record module passed its full verifier (24/24 checks) and test suite (15/15) with the schema migration in place. [POR CONFIRMAR: whether the remaining rollout phases — email confirmation, staff scanner endpoints, final admin panel and hardening — that were still pending as of the last status I have, were completed afterward.]
Evidence from the live site
Real screenshots from the live site and admin tools.
Staff-only login for the record-attempt registration module.
Implementation status: schema migration, bugs found and fixed, 15/15 test suite — point 3 and 4.
Public registration page for the Guinness World Record attempt.
Note: the landing page and the record-form screenshots were captured from an already-logged-in admin session, so a thin WordPress toolbar is visible at the very top edge of those two images.
Short version (portfolio card)
2 custom WordPress plugins, an 80,000-folio pool, QR generation and an AI verification pipeline (Tesseract OCR → Groq LLaMA 3.3 70B) running via cron every 2 minutes — plus a separate Guinness World Record registration module, migrated to fix a duplicate-registration bug and verified with a 50-worker concurrency test. Live in production.
Resume / LinkedIn bullets
- Built and shipped a folio-and-QR registration system for 80,000 units with an unattended AI verification pipeline (OCR + LLM) running every 2 minutes in production.
- Diagnosed a global-vs-per-activity uniqueness bug letting duplicate registrations through a Guinness World Record module; fixed it with a composite unique index and a guarded schema migration (v1.1.0 → v1.1.1).
- Verified concurrency safety with a 50-worker load test against the shared registration path before sign-off, confirming zero duplicate or failed registrations.