Summary
Senior ML / MLOps Engineer with 7+ years of commercial experience building and optimizing production ML systems. Currently working in large-scale fintech, focusing on LLM and speech infrastructure, GPU inference optimization, and cost-efficient model serving. Delivered over $500K in annual infrastructure savings and improved latency/capacity by up to 2× in high-load STT and LLM systems. Previously built and deployed computer vision solutions for industrial quality control.
Work experience
07/2023 - present
Senior ML / MLOps Engineer
T-Bank (ex. Tinkoff)
07/2023 - present
- Improved LLMs serving efficiency by 2× through KV-cache optimization, KV sharing, TensorRT-LLM migration, batching, and request routing improvements.
- Accelerated custom STT models by up to 4× using TensorRT and custom INT8 CUDA kernels.
- Reduced annual infrastructure costs by ~$500K for speech recognition workloads.
- Contributed to optimization and productionization of TTS, STT, and Voice Conversion models.
- Participated in the release of the open-source T-one model:
huggingface.co/t-tech/T-one.
06/2022 - 07/2023
Senior Software Engineer
Mirai Vision
06/2022 - 07/2023
- Early engineer at a computer vision startup, leading development of core systems and helping scale the product to production readiness.
- Led a small engineering team and delivered 5 production-grade ML-based inspection systems in Python for defect detection (PCBs, metal parts, truck frames, rebar, bottles).
- Focused on deployment and optimization of ML pipelines, including running computer vision models on resource-constrained devices (e.g., Raspberry Pi).
04/2021 - 06/2022
Software Engineer
Center for Oil and Gas Technologies
04/2021 - 06/2022
- Developed a modular web-based VR training simulator.
03/2019 - 04/2021
ML Engineer
OFTE
03/2019 - 04/2021
- Developed an ML solution for vehicle classification for usedcarsni.com.
Selected projects
08/2020 - 04/2021
ML Engineer / Team Lead
Eurobot 2021
08/2020 - 04/2021
- Led a 5-member team in preparing for the Eurobot 2021 competition.
- Trained and optimized CNN models for efficient inference on Jetson Nano.
- Gitlab: gitlab.com/inno20_eurobot_full/inno20_eurobot_control.
Achievements
- Speaker at Tinkoff.AI Speech Meetup
- 4 software patents in ML/software systems
- Bronze Medal at “I am a Professional” Olympiad
- Best Innovation Certificate in RoboCup Pacific-Asia 2018