Allen - Vision AI Engineer - Intern

Experiencia demostrada en YOLOv8.

Experiencia demostrada en ONNX.

Experiencia demostrada en OpenCV.

Experiencia demostrada en DeepLab.

Experiencia demostrada en SAM 3-assisted auto-labeling.

Built and deployed a human-in-the-loop PPE detection system for industrial safety: trained YOLOv8s (mAP50 ~0.953, 7 classes), exported to ONNX, running live inference on an Intel RealSense D435i; dashboard requires human confirmation before an evidence photo counts as a flagged violation, with full shift/session tracking.. Designed ZoxiVision QC, a full-stack CV inspection system (React + FastAPI + SQLite + WebSocket): built the dataset capture tool, trained a 7-class YOLOv8 model (mAP50 ~0.928) via Roboflow, and iterated the human-review workflow — lot-based registration, grouped evidence photos with bounding boxes, bulk approve/reject — directly from operator feedback.. Currently building SAM3 Studio, an internal auto-labeling tool: upload photos or video, prompt what to label, extract frames or track objects across a video, then export the labels to train YOLO — the same data-engine problem this role is built around.. Integrating a Deep Robotics quadruped robot with a Jetson Xavier NX; developing on-device AI inference nodes for perception and autonomous behavior on real production hardware, not a toy dataset.. Designed a 4-station Industry 4.0 IoT demo lab (ESP32 + MQTT + Node-RED + InfluxDB + Grafana) for production-monitoring demos.. Maintains internal GitLab repositories; collaborates asynchronously in English with cross-functional, international teams.

Integrated YOLOv8 + DeepLab + a custom CNN for leaf-disease detection on a ground robot operating in real coffee crop fields; validated model outputs under real field conditions and iterated on failure cases — the same ambiguous-case judgment work this role calls for.. Deployed on Jetson Orin (Ubuntu); implemented the SLAM pipeline and configured the ROS2 Nav2 stack (costmaps, path planners, trajectory execution) for autonomous crop-row navigation.. Fused camera, IMU, and depth data for robust localization; managed all software in C++ and Python with reproducible test logs and documented design decisions.

Performed systematic root-cause analysis on electromechanical failures; maintained technical records with full traceability.

Programmed VFDs and performed electromechanical installations; validated technical drawings against specifications and documented configurations for team replication.