Looking for a job

Philip Konokhov

@philipkonokhov

Machine Learning Engineer at Yandex, productionizing diffusion models for autonomous driving and reducing collision rates by 45%.

Russian Federation

Message

At Yandex, I developed and productionized a diffusion model for realistic traffic generation in autonomous driving simulation. I owned its ML pipeline from training and closed-loop evaluation through production integration and safety policies, reducing ADE by 23%, collision rate by 45%, and off-road rate by 55%.

At MTS, I led an ML team developing realistic 3D scene reconstruction and rendering with 3D Gaussian Splatting. I also extended the pipeline to improve the rendering of backgrounds, skies, and distant scene elements.

Across Hoddies, Domclick, and Samsung, I researched generative and 3D vision methods, deployed computer vision models, and worked on visual odometry. My work has included trajectory generation, 3D reconstruction, object detection, and motion representations for 3D rotation estimation.

Experience

Work history, roles, and key accomplishments

Current

Machine Learning Engineer, Self-Driving

Current

Yandex

Jul 2024 - Present (2 years 3 months)

Developed and productionized a diffusion model for realistic traffic generation in autonomous driving simulation. Owned the end-to-end ML pipeline, including model training, closed-loop evaluation, production integration, and safety policies.

Diffusion Models Autonomous Driving ML Pipelines Model Training Evaluation Production Safety Policies Inpainting Trajectory Generation

Current

Data Scientist / Core ML Engineer

Current

Hoddies

Jun 2022 - Present (4 years 4 months)

Researched generative and 3D vision approaches, including 3D object generation, 3D photo inpainting, text-to-video, and text-to-scene generation. Explored novel conditioning methods and ASR approaches for identifying speech-related problems in children.

PyTorch Stable Diffusion ControlNet Generative Models 3D Vision ASR NVIDIA NeMo

ML Team Lead

MTS

Mar 2024 - Jul 2024 (4 months)

Led an ML team developing realistic 3D scene reconstruction and rendering using 3D Gaussian Splatting. Defined research directions, project plans, and technical hypotheses; collected and processed real-world datasets.

Gaussian Splats 3D Reconstruction Team Leadership Research Dataset Curation Rendering

Data Scientist, Chief Development Engineer

Domclick

Jul 2020 - Jun 2022 (1 year 11 months)

Developed and deployed object detection and instance segmentation models for real-estate imagery and built supporting dataset preparation and annotation pipelines. Developed 3D reconstruction components including Structure-from-Motion, single-image room layout reconstruction, and point-cloud alignment.

Object Detection Instance segmentation 3D Reconstruction Structure From Motion Point Clouds Deep Learning Computer Vision

Data Scientist, Development Engineer

Domclick

Dec 2019 - Jul 2020 (7 months)

Developed and deployed a real-time emotion recognition system for analyzing reactions to advertising. Worked on employee scheduling optimization and monocular depth estimation using gradient boosting and deep learning methods.

Emotion Recognition Real Time Systems Gradient Boosting Deep Learning Depth Estimation Optimization

Assistant Engineer, Visual Understanding

Samsung

Aug 2018 - Dec 2019 (1 year 4 months)

Worked on learnable visual odometry based on optical flow; reimplemented and trained PWC-Net in TensorFlow with performance comparable to the original PyTorch implementation. Developed motion representations combining depth and optical flow and implemented quaternion and SO(3) operations for 3D rotation estimation.

Visual Odometry Optical Flow PWC Net TensorFlow PyTorch 3D Rotation Quaternions SO(3)