EDGE NPU DEPLOYED

Rogelio Leonardo Mendez Macias

Embedded Computer Vision & Edge AI Engineer

INT8 Quantization & NPU Deployment (Hailo-8, IMX500) · YOLOv8 · PyTorch · ONNX · ADAS & Autonomous Systems

Published Author (Springer LNCS)

1st Place National Hackathon (IBM-judged)

Falling Walls Lab Finalist

View Projects

HARDWARE PIPELINE METRICS

SYSTEM OPTIMAL

30–58 FPS

Hailo-8 Edge Perception

230 ms

End-to-End Latency

88% INT8

Model Compression

>93% Acc.

Central OCR Precision

Verified in hardware benchmarks & Springer LNCS peer-reviewed paper

PROFILE // BACKGROUND

About Me

Rogelio Leonardo Mendez Macias - Embedded Computer Vision & Edge AI Engineer

Electronics Engineering student at UAM Azcapotzalco (graduating Dec 2026), specializing in embedded computer vision and edge AI for real-time perception systems. My research on on-device vehicle perception was peer-reviewed and published in Springer LNCS at MCPR 2026, and I recently won 1st place in a national AI hackathon judged by IBM. My background in control theory and queueing theory shapes how I design real-time embedded systems, not just how I train models. Currently a finalist at Falling Walls Lab, competing to represent Mexico.

⚡ Specialized in Edge AI, NPU Deployment & Real-time Vision

🏆 Springer LNCS Author & 1st Place National Hackathon Winner

📐 Strong engineering foundation in Control & Queueing Theory

SYSTEM ARCHITECTURE // TOOLCHAIN

Technical Skills

🧠

Edge AI & Model Optimization

YOLOv8 PyTorch PyTorch Geometric OpenCV SegFormer ONNX INT8 Quantization HEF Compilation Hailo-8 NPU Sony IMX500 CUDA

⚙️

Embedded Systems & MLOps

Raspberry Pi 5 (Linux) Docker MQTT REST APIs Frigate NVR Node-RED InfluxDB Grafana

🔬

Systems Fundamentals (differentiator)

Control Theory (BIBO stability, transfer functions) Queueing Theory PID / Ziegler-Nichols tuning

💻

Languages & Tools

Python (Advanced) MATLAB NumPy Pandas SQL LaTeX Jupyter Git/GitHub

APPLIED RESEARCH & EMBEDDED DEPLOYMENTS

Featured Projects

01 / 04

PROJECT 01 // THESIS & SPRINGER

Cognitive ADAS System

Hailo-8 NPU · Distributed Perception · OCR & LLM

Hailo-8 NPU INT8 MQTT ▶ Video Demo

Explore Project & Video →

PROJECT 02 // 1ST PLACE IBM HACKATHON

FloodSense

XGBoost & IBM watsonx / Granite · Real-time Hydrology

XGBoost watsonx Granite ● Live Demo

Explore Project & Demo →

PROJECT 03 // SAMSUNG INNOVATION CAMPUS

AI Road Risk Intelligence

RiskLSTM & YOLOv8 · BDD100K / BDDA · Temporal Modeling

YOLOv8n RiskLSTM PyTorch ▶ Video Demo

Explore Project & Video →

PROJECT 04 // SAMSUNG INNOVATION 2024

IoT Plant-Care System

Dual-MCU & Telemetry · Horticulture Automation

IoT Sensors Microcontrollers Telemetry ▶ Video Demo

Explore Project & Video →

PROJECT 01 // THESIS & SPRINGER PUBLICATION

Cognitive ADAS System (Thesis, UAM Azcapotzalco, 2025-2026)

Problem & Architecture:

Three-layer distributed architecture for vehicle perception — edge perception on Hailo-8 (30-58 FPS), central OCR layer (>93% precision), and an LLM+TTS cognitive layer, coordinated over an MQTT backbone.

Stack & Hardware:

Hailo-8 NPU INT8 quantization OCR LLM TTS MQTT

Key Results:

230ms end-to-end latency, 88% model compression, zero operational failures. This project is the basis of a peer-reviewed publication in Springer LNCS (MCPR 2026) and of a Falling Walls Lab finalist pitch.

Watch MCPR Demo Springer LNCS Paper

Play MCPR 2026 Demo

PROJECT 02 // 1ST PLACE IBM HACKATHON & ADIP INCUBATION

FloodSense

Problem & Architecture:

Real-time hydrological intelligence system built in 48 hours for a national AI hackathon. 3-agent architecture: an XGBoost flood classifier, a risk regressor, and an alert-generation layer powered by IBM watsonx/Granite.

Stack & Hardware:

XGBoost IBM watsonx Granite

Key Results:

ROC-AUC 0.86 (classifier), R²=0.999 (regressor); covers 532 zones across 16 Mexico City boroughs with 6-hour anticipation, trained on 591,706 historical flood records from CONAGUA (1877-2024). Won 1st place (IBM-judged) and was selected for ADIP I+D+i incubation.

Live Demo

PROJECT 03 // SAMSUNG INNOVATION CAMPUS 2025-2026

AI-Powered Road Risk Intelligence (Samsung Innovation Campus, 2025-2026 cohort)

Problem & Architecture:

A 5-stage pipeline for road risk modeling built with a 3-person team split across detection, segmentation, and temporal risk modeling.

Stack & Hardware:

YOLOv8n SegFormer-B0 GCNEncoder RiskLSTM Bayesian Neural Networks / MC Dropout

Key Results:

Individually designed and validated the RiskLSTM strategy — Pearson r=0.916 on 61,345 BDD100K images plus 925 real-world BDDA dashcam clips.

View on GitHub Presentation Video

Play Presentation Video

PROJECT 04 // SAMSUNG INNOVATION CAMPUS 2024 & SANTANDER RETO

IoT Plant-Care System for Horticulture (Samsung Innovation Campus, 2024 cohort)

Problem & Architecture:

An IoT system for horticultural plant care, built with a cross-institutional team including classmates from IPN (Instituto Politécnico Nacional).

Stack & Hardware:

IoT Sensors Microcontrollers Cloud/Telemetry Dashboard Embedded Hardware

Key Results:

Presented in front of Samsung's senior leadership and submitted to the Santander Reto Universitario challenge.

Watch Video

Play Project Video

PEER-REVIEWED SCIENTIFIC RESEARCH

Publications

SPRINGER LNCS · MCPR 2026

Méndez-Macías, R.L., Villegas-Cortez, J., Ferreyra Ramírez, A., Zúñiga-López, A., Cordero-Sánchez, S. "Embedded System for Vehicle Environment Perception and License Plate Recognition (LPR) Using Computer Vision and Deep Learning." In: Pattern Recognition. MCPR 2026, Lecture Notes in Computer Science, vol. 16623. Springer, Cham (2026), pp. 247–258.

DOI: https://doi.org/10.1007/978-3-032-28393-1_22

Presented at MCPR 2026 (18th Mexican Conference on Pattern Recognition, organized by INAOE), Ciudad Juárez, Chihuahua, June 2026.

HONORS // HACKATHONS // CONGRESSES

Awards & Recognition

🏆

Finalist, Falling Walls Lab (Mexico City) — competing to represent Mexico, presenting the ADAS project

🥇

1st Place, Hackathon Concienc.IA 2026 (Young AI Leaders CDMX Hub × Tec de Monterrey CCM, IBM-judged) — FloodSense

🎖️

MUTVI Recognition — UAM International Multidisciplinary Colloquium on Information Visualization, oral presentation "El copiloto que nunca duerme"

🎙️

Presenter, NEO International Congress (2025) — distributed AI architecture and real-time perception pipelines

🌟

Presented IoT project in front of Samsung senior executive leadership — Santander Reto Universitario (2024)

INDUSTRY // TRACK RECORD

Professional Experience

Grupo Modelo (AB InBev)

Data Engineering & Process Digitalization Intern

May – Nov 2023

Built Python + SQL automation pipelines that reduced reporting workload by 5+ hours/week; self-initiated a PPE-detection computer vision prototype for industrial safety, beyond assigned scope.

ACADEMIC BACKGROUND // PROGRAMS