Ilia
Sevostianov

01 / Summary

Computer vision and perception engineer: 5.5+ years in CV and 8 years in robotics. My path runs from UAV and railway-safety R&D, through Lead CV Engineer on an automotive ADAS perception platform, to CV Tech Lead of retail computer vision projects at VkusVill.

I design CV systems top-down: from the business goal and L0–L2 requirements to architecture, hardware and acceptance criteria. I turn research into products that hold up in real-world operation, making structured go / no-go decisions at every stage based on metrics, risk and economics.

02 / Experience

1. Mar 2026 — presentMoscow

   Computer Vision Tech Lead, Offline Retail Robotics

   TechVill (VkusVill tech)

   I lead the “Transparent Store” CV track: camera-based shelf inventory and basket pre-check at self-checkout.

   • Own the project end-to-end: project passport, system requirements (L0–L2), specs, DoR, roadmap, risk register and acceptance checklists.
   • Driving the project to its targets: >98% inventory confidence in a lab plus 5 pilot stores; ≥50% fewer stock gaps; ≥80% less inventory labor; ≥95% basket read accuracy (pilot in progress).
   • Manage external CV vendors: code reviews, reproducibility checks, stage acceptance audits — including NO-GO / hold decisions.

   more

   • Built an in-house metric-learning pipeline for SKU retrieval: closed-set, open-set, gallery-based and few-shot.
   • Researched produce recognition for self-checkout; built a CCTV lens & coverage calculator and a CAPEX model for a typical store rollout.
   • Scoped nearby CV initiatives: anti-shrink, cold chain, café CV, robobarista quality control, inventory robot.
   • Hiring: designed vacancies and interview question banks for CV, active-learning, V&V and model-quality roles; wrote a formal performance review.
   • Attended NRF APAC Singapore 2026 and wrote a comprehensive trip and vendor report.

   metric learning · retail CV · vendor acceptance · CAPEX/ROI

2. Nov 2025 — Feb 2026Moscow

   Computer Vision Engineer, Service Robotics

   Mealty Tech

   • Built Docker infrastructure (CPU/CUDA) and CI for the ROS2 robot workspace; deployed a self-hosted ClearML server and agents.
   • Integrated and tuned Nav2 and SLAM Toolbox until the robot navigated stably in a real environment.
   • Developed 2D/3D segmentation, depth fusion and 3D object tracking on RealSense.

   more

   • Built an active-learning node with SAM3 and a vector database — write-up on Habr ↗.
   • Unified sensor, perception and navigation launch files (RealSense / Orbbec): the full robot pipeline starts in one or two steps.
   • Technical docs (launch, visual markers, DDS) and team AI-dev tooling: Claude Code and Cursor rules, agents, skills.

   ROS2 · Nav2 · SAM3 · RealSense · ClearML

3. Jul 2023 — Nov 2025Naberezhnye Chelny

   Lead Computer Vision Engineer

   KAMA JSC (Atom EV)

   Led the end-to-end perception platform for the vehicle: planning, research, PoC and production rollout.

   • Lane detection: F1 +0.11, fewer lane departures.
   • CUDA image processing across the fleet: ~2× faster inference, CPU freed for other services.
   • Removed redundant transports and introduced docker compose: ~3× faster bring-up. Introduced TDD: >80% coverage.

   more

   • Blind-spot detection MVP on YOLOX with vehicle yaw regression; moved blind-spot and NeuralISP algorithms onto production hardware.
   • MOT3D tracker (EKF): fixed a busy-loop, added multi-camera support, reduced false positives.
   • Started ClearML pipelines: train → val → test → inference → metrics → conversion → deployment.
   • Benchmarked and selected sensors for the production platform; worked with partners and suppliers on specs and bug reviews.

   ADAS · lanes · YOLOX · MOT3D · CUDA · TensorRT

4. Feb 2021 — Jul 2023

   Computer Vision Engineer

   Center for Unmanned Technologies

   • UAV precision landing on LED markers (day and night position and orientation); landing-zone detection.
   • LDWS for an electric bus; 360° bird’s-eye view from multiple cameras.
   • Camera & LiDAR calibration, sensor fusion, quantization and real-time tracking. IEEE and RSCI papers, 1 patent.

   UAV · LDWS · BEV · calibration · fusion

5. Mar 2022 — Dec 2022part-time

   Computer Vision Engineer / Team Lead (team of 3)

   TechTrans

   • Built the prototype of “Guardian” ↗ — a railway worker-safety system that alerts when people cross guarded zones.
   • Developed a ROS2-based simulator that cut the cost of testing and experiments.
   • Detection, tracking, depth & disparity estimation, datasets (ClearML) and project management.

   safety · depth · simulation · team lead

6. 2018 — 2020

   Earlier

   • Innopolis University Robotics Lab, intern (2019–20): ball localization for a UR10 arm with a RealSense D435i.
   • JBL Robotics, engineering assistant (2018–19): ROS nodes for a barista robot, gripper and cup-stand design.

03 / Core competencies

04 / Products & open source

Independent products and freelance since 2020 — shipped and monetized.

05 / Achievements & publications

• 2020🥇 1st place, Aerobot 2020 drone competition — line following and landing-pad detection on ROS.
• IEEETripteron stiffness — ieeexplore 9290208 ↗
• AIP · RSCIVertical walking robot simulation (AIP), RSCI papers and 1 patent.
• 2021Innopolis University award for contribution to science.
• 2020Innopolis University award for academic excellence.
• 2018Best student of the Robotics Department, Bauman MSTU.

06 / Education

07 / Contact