Design and maintain containerized ML Infra pipeline using ClearML & Triton Inference Server - reducing e2e PyTorch model runtimes from 2 minutes (per 1 minute video) down to 20 seconds

Scale up ML Inference workflow using Apache Airflow for maximum visibility and scalability

Convert open-source ML models to TensorRT for optimal performance

Improved segmentation codebase as part of efforts to open‐source & publish results on explainability of vision models

Setup & reported on 300+ experiments; established baselines, and reproduced results on other datasets using in‐house model, resulting in CVPR 2023 (Highlight) publication

Developed and deployed driving mode detection PoC onto Raspberry Pi, connecting to fleet of devices using AWS IoT

Assisted chief data scientist to productize deep learning-based lane detection module—TF Serving, ONNX, Apache Airflow

Introduced CI/CD pipelines to validate model integrity and accuracy before deployment to production