Curriculum Vitae

Filed 3 patent applications and owned dataset/model licensing compliance — including identifying a licensing issue tainting training data and derived models, and driving a full re-annotation remediation.

Designed a multi-camera 3D pose auto-annotation pipeline: cross-camera association, uncertainty-weighted triangulation in JAX, and optimisation-based temporal filtering. Over 95% of frames met or exceeded human annotation quality on many sequences.

Patched YOLO26 and RTMO to expose per-keypoint uncertainty at inference; calibrated raw outputs to pixel-scale errors for downstream 3D reconstruction.

Built an internal MLOps platform over PostgreSQL — deterministic, cached, distributed component execution powering annotation, evaluation, and training-data pipelines.

Owned pose-model training end-to-end: custom dataset loaders, augmentations, and fine-tuning for production conditions.

Built full-stack internal tools (Python + Svelte): a multi-camera ArUco calibration tool (sub-cm accuracy at 5–10 m, patent filed) and a microservice orchestration manager UI.

Co-maintained core infrastructure as one of two go-to engineers: PostgreSQL, Docker, GitHub Actions, Grafana, and MLflow.

Tracked CV/ML literature and presented internal deep-dives on pose estimation, tracking, and monocular 3D methods.

Edit layperson summaries of peer-reviewed research for clarity and style.

Pivoted from pure to applied mathematics ~2 years in. Thesis on low-rank tensor methods, randomized linear algebra, and applications to machine learning.

Published 5 papers; 3 accompanied by open-source Python libraries (computational algebra, numerical linear algebra, and machine learning).

Taught 3 courses per year as an assistant, receiving consistent positive feedback from students.