Aditya Patil

Developed unsupervised ML model to detect scaled vandalism/fraud attacks on Google Maps, achieving > 90% precision with near real-time latency. Improved manual labeling throughput by 8x through creation of dedicated tooling and optimized workflow for evaluating user contribution clusters. Implemented semi-supervised graph-based ML pipeline for fraud detection in user-generated content, increasing precision without compromising recall. Engineered algorithms for EU DSA compliance in maps moderation, ensuring verdict stability and data freshness. Optimized database performance and queries, enhancing efficiency across Google Maps teams.

Implemented GPU-optimized algorithms, achieving up to 100x speedup over CPU implementations. Enhanced Lidar SCAM pipeline performance, improving frame rate from < 10 FPS to > 60 FPS. Developed deep learning model for road boundary detection from Lidar data, incorporating unsupervised clustering for preprocessing. Created synthetic dataset generation pipeline in 3D simulation software, improving computer vision model performance by 5% through fine-tuning. Designed Power BI dashboards for support staff, automating weekly reports and individual metric tracking to boost productivity.

Implemented graph convolution layers and developed proof-of-concept model for 3D model generation and variation.

Developed high-performance C application for low-latency CPU usage statistics collection, aiding developers and QA teams in identifying performance issues.

Teaching Assistant for Data Science course.

Technical Head for departmental committees.