Alvaro G.

I design, build and own computer vision systems that run in production on edge devices (like NVIDIA Jetson), on GPU servers in the cloud (AWS, Azure, Google Cloud), or split between both. 15+ years in computer vision, 25,000+ hours on Upwork, 100% Job Success. I own systems end to end: from figuring out what the business actually needs, through data, models and pipelines, to deployment, monitoring, and fixing things when they break at 3am. Several times I’ve been the only CV person on the team, or the first engineering hire, so I’m used to owning the whole thing. How I work: AI-native, human-accountable AI agents are part of my daily workflow. I use agents to write and refactor code, generate tests, dig through unfamiliar codebases, and prototype ideas in hours instead of days. That means I deliver faster, and I can cover more of the stack myself. What doesn’t change: I design the architecture, review every line that ships, and I’m accountable for the result. I only use the tools your security policy allows, and your code and data stay where you want them. Now that code is no longer the bottleneck, architecture decisions matter more than ever. I’ve spent years working through scalability problems, and I’ve taken systems from an idea to POC to production at scale. Real-time systems fail in predictable ways: dropped frames, memory leaks, latency that creeps up under load, models that drift in the field. I’ve hit most of them, and I design so they don’t happen, and you don’t waste time and money fixing them. What I build Edge AI • NVIDIA DeepStream pipelines, custom GStreamer plugins, multi-camera setups, object classification and tracking • Jetson deployment (Orin, Xavier, Nano) with TensorRT optimization (FP16/INT8) under real power and thermal limits • I keep my own Jetson devices for prototyping and testing, so I can reproduce hardware issues without waiting on your setup Cloud • Scalable video and image processing on AWS, Azure and Google Cloud • Streaming footage securely from cameras to the cloud for real-time processing • Computer vision APIs and services at scale: OCR, document reading, image analysis • Keeping GPU costs under control as volume grows Models • Object detection, segmentation, tracking, action recognition, OCR, face recognition • Training on messy real-world data, with evaluation that shows whether a new model is actually better • Retraining pipelines and CI/CD for models, so they keep improving automatically Recent work • Edge AI team, warehouse automation company: real-time detection of pallet movements and cargo localization, integrated with warehouse management systems. Core system in C++, with supporting Python services. • Machine vision lead: real-time intrusion and suspicious-behavior alerting for CCTV, with cameras from multiple clients streaming to AWS GPU servers. Built in Python. • Football player tracking: performance statistics and automatic highlight clips of each player's best moments in a game. Python, Windows desktop app. • License plate recognition from a camera on a moving vehicle: reading plates, classifying their origin, detecting whether the target car is moving, and sending real-time data to a server for decisions. • First engineering hire at an AI startup: owned the system from prototype to production, building data pipelines and analytics for multiple clients. Stack: Python, C++, C, Go · PyTorch, TensorFlow, YOLO, OpenCV, ONNX · DeepStream, TensorRT, CUDA · GStreamer, FFmpeg, RTSP · AWS, Azure, Google Cloud, Docker, Kubernetes, Triton Industries: logistics and warehouse automation, CCTV/security, traffic and license plate recognition, sports video, video advertising, document processing, construction mapping Working with me: Based in Peru (UTC-5), with full overlap with US hours, and I've worked closely with teams all over the world. Fully remote for 13 years.