At Siemens, I built an AI-powered coding assistant for Capital plugin development, combining Continue with an Azure-hosted agentic backend that accesses indexed codebases, documentation, and tools. It improved developer productivity by 47%.
I also extended CVAT for LiDAR and LiDAR-camera calibration, SLAM and Gaussian-splatting 3D scene generation, and graphical annotation workflows for neural-network training data. My automatic extrinsic LiDAR-camera calibration raised accuracy from 35% to 85%, matching human performance and removing manual calibration needs.
Alongside production AI infrastructure, I build from-scratch and systems-focused ML projects, including a Python and C++ implementation of GPT-OSS, CPU Llama inference in standard C++, and reinforcement-learning systems that achieved superhuman Connect4 performance.
