At Planner 5D, I audit computer vision datasets and define standards for their structure, metadata, access, and versioning. I also design labeling taxonomies and annotation guidelines to support consistent, reproducible work.
At Toloka, I architected end-to-end RLHF and SFT data pipelines for foundation model training and evaluation, including complex coding workflows. I developed multimodal preprocessing and LLM-based generation and validation components to reduce manual effort and labeling costs.
At ChemTech AI, I designed and implemented core parts of an AI SaaS platform for manufacturing, from data ingestion and ML inference to factory integrations and deployment. The platform generated $500k–$1.6M in additional monthly profit, and deployed models increased equipment utilization to 96%.
Earlier, I built a workforce scheduling optimization system at NOVARDIS Consulting and designed a predictive ML quality-control system at JACOBS DOUWE EGBERTS that saved approximately $200K annually. My work connects AI and automation to practical operating needs, including factory environments with limited or no internet connectivity.
