AEVO — Agentic Edge-Vision Optimizer
An autonomous MLOps pipeline that trains a YOLOv8 defect-detection model, uses the Claude API to self-tune hyperparameters in real time, then compresses it to 3.6 MB at 122.9 FPS for edge deployment — no human in the loop.
View Project
Monocular SLAM Benchmark — 8 Systems, One Comparison
An independent, hardware-matched benchmark of 8 monocular SLAM/VO systems — ORB-SLAM3, DSO, DROID-SLAM, DPVO, MonoGS, Splat-SLAM, HI-SLAM2, GigaSLAM — scored head-to-head on accuracy, metric scale, robustness, and 3D reconstruction quality.
View ProjectConfidential — AI & Computer Vision for Sports
An end-to-end AI and computer-vision system for a sports-technology client: data annotation, a full training pipeline, tightly coupled AI–CV inference, model optimization, and 3D visualization of results. Specifics are protected under a non-disclosure agreement.
Protected under NDAConfidential — AI Defect Detection for Manufacturing
An AI inspection system for a manufacturing client — including custom hardware for data collection and annotation, model training and testing across multiple defect types, and dimensional analysis. Under a confidentiality agreement, so the details stay private.
Protected under NDA