Hello,
I'm Yuri Mayorov

• Implemented backend using Axum (Rust) for high-performance data processing, FastAPI (Python) for ML services, Gin (Go) and Express (TypeScript) for other components. Developed unified components for ML model inference in NVIDIA Triton (ONNX Runtime). Integrated observability stacks (logging, metrics, tracing) and resilience patterns (circuit breaker, retry policies).
• Built frontend on React, including interactive dashboards for real-time flow data visualization, tools for exploratory data analysis (correlation heatmaps, time series decomposition, feature importance visualization – plotly, highcharts, d3), interface for training new and fine-tuning existing ML models.
• Wrote baseline ML models, primarily using CatBoost, XGBoost and PyTorch, many remained in production due to better speed-to-accuracy ratio. Implemented adaptive retraining system with automatic drift detection. Developed physics-based checks for input data consistency and anomaly detection.

Implementation of ML Driven Virtual Flow Metering instead of traditional physical flow meters costing $500k+ saves up to $3 million annually on one major platform through reduced equipment maintenance costs, decreased downtime, and increased production planning accuracy.

Flow Simulator

Multiphase flow simulator for modeling hydrodynamic processes based on specified flow rate parameters, PVT properties, system geometry and other parameters (analog to commercial solutions Schlumberger PIPESIM and OLGA).

• Implemented solver components in Rust and cross-platform GUI in React that calls the solver. Developed application architecture and build pipeline that automatically compile a single codebase into web version on Express and desktop application on Electron. Implemented batch execution algorithm for multiple calculation cases based on recombinant parameters.
• Implemented system for rendering calculation results directly on well geometry as color maps and charts along the trajectory. Developed complex UI components for simulation parameter input: drag-and-drop, PVT table editors, interactive forms for various parameter input, data validation and import from Excel/CSV.

Development enabled partial abandonment of expensive commercial licenses for similar software, saving ~$2 million per year.

Other Projects

Within other projects, worked with TensorFlow, OpenCV, Three.js, D3.js technologies, participating in the development of geological data intelligent analysis system with automatic seismic interpretation based on transformer models (GeoMind), well placement optimization service, and field development planning system.