$ cat work/timeline.log | tail -n 4

Work.

Four roles, one through-line: ship CV systems that survive the field. Multimodal sensors, edge constraints, domain shift.

  1. Machine Learning Engineer

    › TransferX

    Feb 2026 — present
    • ›Designed PostgreSQL schema and ETL pipelines for EdTech enrollment and credit tracking.
    • ›Built ML prototypes for university classification and ranking on noisy, real-world data.
    --postgresql--etl--python

  2. AI Systems Integration Engineer

    › Alta Potentia LLC

    Mar 2025 — Nov 2025
    • ›Deployed YOLOv8 on NVIDIA Jetson at sub-100ms latency for real-time object detection in retail and security applications.
    • ›Quantized models with TensorRT, achieving 60% size reduction while maintaining accuracy, enabling edge deployments that reduced infrastructure costs by 30% per device.
    • ›Built data pipeline processing 500K+ frames in coordination with clients, reducing manual review workload by 16% in retail loss prevention.
    --yolov8--jetson--tensorrt--onnx--vision applications

  3. ML Engineer Intern

    › OWL Autonomous Imaging

    Jan 2024 — May 2024
    • ›Extended MGNet (originally monocular RGB) into a dual-encoder thermal+RGB fusion model for real-time panoptic segmentation and self-supervised monocular depth estimation; shipped in OWL's downloadable ADAS perception software and anchored the 3D scene-understanding roadmap.
    • ›Architected the multimodal data pipeline for OWL's thermal+RGB+LiDAR sensor stack: intrinsic/extrinsic calibration, temporal sync, RGB-thermal parallax correction, Detectron2 panoptic auto-labeling, and 2D-to-3D label projection into LiDAR; lifted panoptic segmentation mIoU 20% on the production Faster R-CNN + panoptic pipeline.
    • ›Built an interactive panoptic annotation UI (sliders, color-coded class picker, parallax-aware overlay) over Detectron2 auto-labels; accelerated labeling 40% across 50K+ frames and cut third-party annotation spend.
    • ›Built a YOLOv5 data-quality layer for automated label QA, class balancing, and parallax-aware filtering; lifted Faster R-CNN detection mAP 15%.
    • ›Engineered real-time super-resolution and denoising for 16-bit thermal+RGB streams (1280x800 at 30 FPS); Dockerized into the CI/CD pipeline for on-vehicle inference on Jetson Orin.
    --detectron2--yolov5--thermal-rgb--panoptic--docker--lidar--jetson-orin

  4. Machine Learning Engineer

    › WhiteHat Education Technology

    Jun 2020 — Aug 2021
    • ›Built robust unit testing frameworks (PyTest) for 300+ Python ML projects, drastically reducing production failures.
    • ›Deployed automated data drift detection (MLflow, Airflow) to ensure stable model performance across diverse datasets.
    • ›Analyzed marketing, sales, and student data, implementing algorithms that boosted lead conversion, revenue, customer retention and learner engagement.
    • ›Led a seven-person team to design data-driven A/B tests and predictive analytics workflows, significantly improving student success rates and overall program efficiency.
    --mlflow--airflow--team-lead--production-ml--founding-team