Luis Fernando Vargas Artiaga

Senior AI Engineer

GovDash

Apr 2022 - Current (4 years 2 months)

• Fine-tuned Meta's LLaMA 3.2 Vision model (11 B parameters) to improve its understanding of medical images, enhancing diagnostic accuracy. • Utilized the Radiology_mini dataset, containing chest X-rays and expert-annotated radiology reports, to train the model for detecting abnormalities. • Applied LoRA (Low-Rank Adaptation) for parameter-efficient fine-tuning, optimizing performance while reducing memory and compute requirements. • Achieved significant improvements in disease identification from X-rays, supporting AI-assisted diagnostic workflows and enhancing radiologist decision-making.

AI/ML Engineer

Kitrum

Sep 2021 - Mar 2022 (7 months)

• Designed and implemented an INT16-only quantization solution for YOLO models on specialized XPU hardware lacking floating-point operations, reducing model size by 70% while maintaining similar accuracy. • Developed an adaptive scaling algorithm for weight and bias quantization, minimizing memory footprint and enabling deployment on embedded and low-power AI hardware. • Ensured efficient, high-speed inference on XPU hardware by eliminating reliance on CPU processing, achieving a 3x speedup in inference time, significantly improving real-time AI applications. • Applied model compression and pruning techniques, optimizing computational complexity and improving inference speeds by 65%, ensuring high performance on edge AI hardware. • Col

G

AI Engineer

Grab Holdings

Apr 2020 - Sep 2021 (1 year 6 months)

• Designed and implemented an AI-powered Driver Guidance System for Car Washes, reducing inefficiencies and potential damage risks by 13%, ensuring smoother and safer vehicle alignment onto automated conveyor belts. • Developed and fine-tuned YOLO and SSD models for accurate real-time detection of cars and tires, integrating them into a DeepStream-based Jetson Nano edge computing system, enabling inference at 50ms per frame. • Optimized model inference using TensorRT, reducing latency by 15%, leading to improved system responsiveness and enhanced customer experience. • Engineered an algorithm for real-time vehicle positioning on conveyor belts, increasing automation efficiency, reducing manual corrections, and improving throughput at bu

D

AI Developer

Doodle Labs

Apr 2014 - Mar 2020 (6 years)

• Automated the detection and localization of electrical poles in urban and rural environments using deep learning and Google Street View imagery, achieving 90% accuracy, significantly improving mapping precision and infrastructure planning. • Applied photogrammetry techniques to enhance localization accuracy, improving mapping precision by 5–10 meters, benefiting electrical grid maintenance and expansion projects. • Leveraged CUDA acceleration to optimize a processing pipeline for over 1 million images, achieving a 60% reduction in processing time, enabling large-scale dataset analysis efficiently. • Integrated the AI-powered detection and mapping solution into a GIS-based system, streamlining post-storm damage assessments and accelera

Luis Fernando Vargas Artiaga

About

Detail

Timeline

Résumé