Mahesh Dhokane

I’m a result-driven Quality Analyst and Data Annotator with nearly 3 years of high-impact, onsite experience at NVIDIA (via Randstad Technologies). I specialize in high-precision autonomous vehicle (AV) datasets that strengthen machine learning perception through immaculate visual data training.

In my role, I annotated real-world autonomous driving datasets using the NVIDIA 3DGT Ground Truth tool, synchronizing multi-camera frames and LiDAR point-clouds. I label dynamic road obstacles—vehicles, pedestrians, VRUs, cyclists, and animals—with exceptional geometric and rotational accuracy (within 10–15 cm).

I’m especially strong in resolving complex data discrepancies such as sensor misalignment, merge conflicts, and temporal object propagation errors. I also apply class-specific labeling protocols for non-standard vehicles and independently isolate rotating sub-components (doors, attached cargo, and trailers), including parent-child vehicle structures.

To keep dataset integrity at a rigorous standard, I run extensive QA/QC audits, content moderation, video annotation workflows, and transcription reviews. I’ve also streamlined processing using Hebe optimization for view rotation and focal zooming, and I contribute to projects like Dynamic Obstacle Labeling by recalibrating 3D/2D projection models to align camera streams with LiDAR returns.