Open to opportunities

Михаил Котюшев

@0005118

Message

I build computer vision and ML systems from research prototypes to optimized production inference.

Kazakhstan

Message

At GeoAlert, I extend SAM 3 for Mapflow, build interactive TensorRT and Triton inference services, and improve tree-crown instance segmentation while cutting model size by 50%. Previously at Faceplate Industrial AI, I led AI integration for an electrical-energy client, delivering SCADA incident-summary pipelines and power-grid forecasts with MAPE below 10% long-term and 5% short-term.

I've also taken Wi-Fi R&D from research to shipping products at Huawei, built embedded hardware and automated systems, and earned Kaggle Expert status through solo vision, graph, and time-series competition results. My work spans PyTorch, model compression, LLM systems, C++ simulation, and production deployment.

Experience

Work history, roles, and key accomplishments

Current

Senior Research Engineer

Current

GeoAlert

Feb 2026 - Present (6 months)

Extended SAM 3 with polygon prompting and improved tree-crown segmentation. Built interactive inference service with TensorRT and Triton, and compressed ViT-L models for efficient deployment.

SAM 3 TensorRT Triton Onnx ViT Distillation Pruning Few Shot Learning Segmentation Object Detection

Machine Learning Engineer

Faceplate Industrial AI

Feb 2024 - Feb 2026 (2 years)

Led the company's AI function, delivering AI and ML integration for a major electrical-energy client. Built an LLM, RAG and human-in-the-loop pipeline converting high-volume SCADA alerts into incident summaries and action plans, and designed automated offline evaluation of summary and action quality.

LLM RAG Human In The Loop SCADA XGBoost Prophet scikit learn Scipy VLLM Docker MCP REST Time series Forecasting Evaluation

Research Engineer

Huawei

Jun 2019 - Feb 2023 (3 years 8 months)

Conducted Wi-Fi R&D across five research projects, two of which were deployed in shipping products. Developed adaptive packet aggregation, predictive MIMO pre-coding, and multi-link load balancing, and built a testbed for automated experiments.