Quito, Ecuador · Remote · UTC−5 · Open to international roles
AI & XR Engineer. I ship machine learning to production, from on-device detection for a Stanford project to retrieval systems that cite their sources.
0.96
mAP50, detection (validation) · on-device tuna detector, Stanford
49 ms
inference per image on a 2-core server
14
h-index · 640 citations · 30+ papers
2016
building XR in Unity since
From a robotics simulator in 2016 to on-device detection in 2026. Photos from the original projects.
I start studying artificial intelligence at ESPE.
First paper: a real-time Unity3D–MatLab simulator for robotics. 88 citations, still my most cited.
First place at the Galardones Nacionales, awarded by SENESCYT.
I found Inmersoft and register a software patent.
At Bagó I build interactive and AI tools, such as an AI-assisted platform for the product managers' creative sessions.
Stanford: on-device detection for Buscapez.
Machine learning in production, retrieval systems and the tools I use every day.
Featured · Stanford project · 2026
An on-device tuna detector built for a Stanford project. A 4-class YOLO model trained in PyTorch runs fully offline on iOS and Android through ExecuTorch, and the same model is served with ONNX Runtime on the server. Detection is very precise; species classification keeps improving as the team validates more real photos.
In development
Retrieval · 2026
A retrieval assistant that answers questions about my work and cites its public sources. Hybrid search over pgvector, answers written by Claude, with a daily budget and logs.
Live demo: Ask WashingtonRepository: Ask Washington (GitHub)
Generative AI platform
A self-hosted, invite-only platform that turns local LLM and ComfyUI models into configurable services, with a Redis job queue and real-time progress over WebSockets.
Agent tooling
A live 3D office, on the web and on iPhone, that shows what every Claude Code agent is doing across machines, with permission approval and orchestrator-led teams.
Computer vision
Gaze tracking with an ordinary webcam: an ETH-XGaze CNN, per-session screen calibration and a click-through desktop overlay, with no infrared hardware. Accuracy: 3.97° overall (2.51° in the central region).
This is the Buscapez model: a YOLO detector for four tuna species, exported to ONNX. Pick an example or use your own photo.
Runs in your browser. Your photo is never uploaded.
Example photos: Tuna Fish Detection by Washington X. Quevedo, Roboflow Universe, CC BY 4.0. Model: Ultralytics YOLO26n, AGPL-3.0.
or drop an image on the frame
Virtual and augmented reality for training, robotics and heritage, built in Unity.
Running through a virtual maze on a real treadmill.
Driving a remote robot from an immersive interface with haptic feedback.
VR training to recognize and assemble engine components.
Virtual reconstruction of historic buildings and their surroundings.
Virtual instrumentation of a multirotor drone with robotic arms.
From a P&ID diagram to an interactive 3D process plant, operated with the hands.
30+ peer-reviewed papers on virtual reality, robotics and education technology, published by Springer, ACM and IEEE. h-index 14, 640 citations.
2016 · 88 citations
International Conference on Augmented Reality, Virtual Reality and Computer Graphics (AVR, Springer LNCS)
2018 · 55 citations
International Conference on Augmented Reality, Virtual Reality and Computer Graphics (AVR, Springer LNCS)
2017 · 54 citations
IT Convergence and Security 2017 (Springer LNEE)
Ask Washington answers questions about my experience using only public sources: my CV, papers and repositories. Every answer cites where it comes from.
Answers are generated from public sources and may contain errors.
Connecting to Ask Washington
The chat could not load.
Open Ask Washington in a new tab