Case Studies

Context, technical decisions and measurable results

AI-powered OCR system for credential validation

Context

Need to automatically process physical credentials with field validation, orientation and data extraction. Manual process was slow and error-prone, requiring constant human validation.

Technical Decision

Implemented a system with YOLOv8 for field detection through computer vision, OpenCV for orientation validation and optimized Tesseract OCR. Hardware control via RS-232/USB-COM with device reverse engineering.

Result

35% reduction in OCR errors, processing in less than 2 seconds per credential. Fully automated system that eliminates the need for manual validation in most cases.

Reducción de errores: 35%Tiempo de procesamiento: <2 segundos

PythonYOLOv8OpenCVTesseractPyTorchSelenium

Gamified educational platform

Context

Educational platform that needed to support over 15,000 concurrent users with Q&A system, authentication, profile management, lives system and virtual wallet. Required high availability and low latency.

Technical Decision

Modular architecture with NestJS using decoupled modules, Prisma ORM for database management, strategic caching with Redis, and complete CI/CD pipeline with CodeBuild, ECR and Load Balancer on AWS. Implemented BFF (Backend for Frontend) with Redux slices.

Result

Architecture supporting +15,000 concurrent users with p99 <180ms. CI/CD pipeline reduced delivery times by 40%. Modular system facilitates maintenance and scalability.

Usuarios concurrentes: +15,000Latency p99: <180msReducción tiempos entrega: 40%

NestJSDockerPostgreSQLPrismaNode 22AWSEBS3Redis

Electoral management platform

Context

System for electoral campaigns that required real-time facial recognition during mass events, hierarchical tree of affiliates, geofences and interactive maps with growth KPIs by geographic zone.

Technical Decision

Implemented live facial recognition using AWS Recognition with Kinesis for video streaming, optimized hierarchical tree with efficient data structures, and geofences with real-time KPI calculation. Bulk user upload system with validation.

Result

Real-time facial recognition during events with video stream processing. Hierarchical tree enables dynamic org chart visualization. Geofences and maps provide real-time growth KPIs by zone.

Procesamiento en tiempo realÁrbol jerárquico escalableKPIs geográficos en tiempo real

CakePHP 3MySQLAWS RecognitionKinesisTreesPHP

← Back to projects