Vladimir Kulyashov

About

Russian

native

English

C1

Experience

1. Nov 2025 – present

   MLOps / LLM Ops Engineer

   GNIVC, Moscow

   I develop the GNIVC AI Sandbox, the company's internal environment for AI experiments and applied AI services (1,000+ users). I'm responsible for model inference in the Federal Tax Service of Russia's pilot and production environments, and I take part in designing the FTS AI Platform.

   • Deployed and run an LLM inference service with an OpenAI-compatible API on vLLM, SGLang and NVIDIA Dynamo (Qwen 3.x, DeepSeek V4 and others). Moved production models between engines and schedulers with no loss of availability; fixed degradation under load and stuck queues.
   • Optimise inference: continuous batching, prefix and FP8 KV cache, queue policies and overload protection. Median TTFT of the key model improved about 3x (~1.1 → 0.4 s), and FP8 nearly doubled KV cache capacity.
   • Deploy, update and maintain models on inference servers in the FTS pilot and production environments; benchmark models for speed and cognitive ability.
   • Take part in designing the FTS AI Platform: concept and design documents, model runtime architecture (Prefill/Decode disaggregation), GPU selection and allocation, resource requirements.
   • Designed the sandbox Kubernetes cluster with a GPU node pool (NVIDIA H100, H200, RTX 6000 Ada), GPU Operator, KAI Scheduler and Grove, Longhorn, Calico, MetalLB and Envoy Gateway; automated node configuration with Ansible.
   • Deploy and upgrade sandbox services without downtime: Open WebUI, GPUStack, JupyterHub, Qdrant, a RAG stack, MinIO, ClearML, Langfuse, BI, n8n automations. Upgraded Langfuse from 3.x to 4.x, migrating ~1.9M events with no data loss.
   • Selected and deployed a decision model in the sandbox: compared open options (Kev, Laya, Clef and others) on one task set and chose decisions 3.0.
   • Built the CI/CD and GitOps setup: Argo CD with Keycloak SSO, image builds on Argo Workflows and kaniko into Artifactory, Bitbucket events via Argo Events, Vault with External Secrets Operator, Kyverno policies.
   • Own platform security: hardened images (vLLM, SGLang, DCGM Exporter and others) to zero fixable CRITICAL/HIGH findings in Trivy, restricted pod profiles, network policies, work with the security team.
   • Run observability and operations: VictoriaMetrics, Grafana, Loki, Vector, a service status page, incident reviews, capacity planning for GPU, memory and storage.
   • Built Model Router, an in-house HTTP router for vLLM/SGLang with load balancing, failover, retries and metrics.
   • Support the sandbox handover: preparing for acceptance and migration to another site, deployment guides for the operating organisation, user support.

   • Kubernetes
   • NVIDIA GPU Operator
   • KAI Scheduler
   • Grove
   • NVIDIA Dynamo
   • vLLM
   • SGLang
   • GPUStack
   • Helm
   • Argo CD
   • Argo Workflows
   • Argo Events
   • Vault
   • External Secrets
   • Kyverno
   • Keycloak
   • Longhorn
   • Calico
   • MetalLB
   • Envoy Gateway
   • VictoriaMetrics
   • Grafana
   • Loki
   • Vector
   • n8n
   • Langfuse
   • ClearML
   • MinIO
   • Qdrant
   • PostgreSQL
   • Ansible
   • Trivy
   • Go
   • Python

2. Dec 2023 – Oct 2025

   MLOps Engineer

   Rusagro Group, Moscow

   Developed MLOps and DevOps practices in the ML & AI division, supporting 3 teams and the infrastructure for the full model lifecycle.

   • Moved ML applications to Docker, laying the groundwork for Kubernetes.
   • Set up two Kubernetes dev clusters (1 and 6 nodes) for computer vision and classic ML services.
   • Built GitLab CI templates (build, tests, vulnerability scanning, deploy), cutting pilot launches from months to a few weeks.
   • Introduced SAST and container scanning with Trivy.
   • Deployed the core MLOps stack: ClearML, MinIO (S3), HashiCorp Vault, Prometheus, Grafana, Loki, Alloy.
   • Configured infrastructure with Ansible together with DevOps engineers and the infrastructure team.
   • Built and maintained an internal utility library that standardised ML development.
   • Sat on architecture committees, reviewed solution designs and code, and wrote recommendations on MLOps practices.

   • Kubernetes
   • Docker
   • Helm
   • GitLab CI
   • Trivy
   • ClearML
   • MinIO
   • Vault
   • Prometheus
   • Grafana
   • Loki
   • Ansible
   • PostgreSQL
   • MongoDB
   • Python
   • Bash

3. Apr 2023 – Dec 2023

   Senior Computer Vision Engineer

   Rusagro Group, Moscow

   Development automation and architecture of CV services.

   • Built and maintained an internal computer vision library.
   • Made architecture decisions for CV services and helped plan new projects.
   • Deployed experiment tracking.
   • Designed a storage and catalogue system (DWH) for labelled and unlabelled data.

   • Python
   • PyTorch
   • OpenCV
   • ClearML
   • Docker

4. May 2022 – Apr 2023

   Computer Vision Engineer

   Rusagro Group, Moscow

   Projects for the meat business — counting livestock and spotting sick or dead animals.

   • Together with the team, brought pig counting accuracy in pens to 99+%.
   • Built data collection pipelines and trained models.

   • Python
   • PyTorch
   • OpenCV
   • TensorRT

5. Jul 2021 – May 2022

   Lead ML Engineer

   Forpost, Perm

   Video analytics, shipping models to production and leading a group of ML engineers.

   • Set up and ran cloud infrastructure for neural network training.
   • Wrote custom layers and implemented algorithms from recent computer vision papers.
   • Standardised ML pipelines and shipped models to production.
   • Hired, mentored and led ML engineers, set and tracked tasks, and handled technical talks with clients.

   • Python
   • TensorFlow
   • PyTorch
   • Darknet
   • OpenCV
   • TensorRT
   • TF Serving
   • GCP
   • AWS

6. Aug 2020 – Jul 2021

   ML Engineer

   Forpost, Perm

   Detection, segmentation and classification models for video analytics.

   • Collected and labelled data, fine-tuned and improved models, prepared them for production.
   • Wrote tools and scripts for data collection and processing.

   • Python
   • TensorFlow
   • Darknet
   • OpenCV

7. Mar 2020 – Aug 2020

   Product Analyst, Dom.ru Movix

   ER-Telecom Holding, Perm

   Product analytics for the Movix TV service.

   • Tested hypotheses, computed product metrics, segmented the customer base.
   • Analysed user search queries and subscriber traffic.
   • Designed and populated the analytics team's database and automated reporting.

   • Python
   • SQL
   • pandas
   • Statistics

Teaching

1. Sep 2024 – Jun 2025

   MLOps Lecturer

   Perm State University, Perm

   Taught a master's course on MLOps and integrating machine learning into web applications.

   • Designed lectures and labs on automating the model lifecycle: data, training, experiment tracking, deployment.
   • Supervised the capstone: from data collection to a model behind an API with performance and quality metrics.
   • Taught students tools for automating and monitoring ML projects, including experiment tracking and CI/CD.

   • Python
   • Docker
   • FastAPI
   • CI/CD
   • MLOps

Projects

Skills

Education

• 2021Neural networks in computer vision, Perm State University, Regional Institute of Continuing Education
• 2020Data Scientist program, Skillbox
• 2020Professional Scrum Developer, Agilix consulting
• 2019mlcourse.ai — machine learning, Open Data Science
• 2008Real estate appraisal and management, Perm State Technical University