Positioning
Infrastructure / Platform Lead who operates private AI / GPU inference as production infrastructure, on a deep SRE base: private cloud, distributed storage, highload traffic platforms, DDoS resilience, observability and network telemetry. The differentiator is the fusion — an infrastructure engineer running LLM / RAG inference as an operated platform used by real engineering teams. Best-fit roles: AI / GPU Infrastructure Lead, Inference / AI Platform Lead, Head of Infrastructure, Principal / Staff SRE.
Summary
8 years in SRE and infrastructure engineering. Strongest areas: Ceph and private cloud, highload infrastructure, P0/P1 incident response, multi-tenant observability, BGP/BMP telemetry, IaC, platform migrations and on-prem LLM/RAG systems for SRE workflows.
Core skills
- AI infrastructure: private LLM serving, GPU inference, vLLM, llama.cpp, LiteLLM routing, Qdrant, bge-m3, reranking, RAG, Langfuse
- Private cloud: Proxmox, OpenStack, VMware, Ceph, ZFS
- IaC: OpenTofu, Terraform, Packer, Ansible, GitLab CI
- Observability: VictoriaMetrics, VictoriaLogs, ClickHouse, Grafana, Vector
- Network: BGP, BMP, SRv6, SR Policy, EVPN, FlowSpec, Juniper, Huawei
- Security & access: Vault/OpenBao, RADIUS, Wazuh, audit trails
- Programming: Go, Python, Bash, FastAPI, Flask, Celery
- Data & messaging: PostgreSQL, Redis, NATS JetStream, ClickHouse
Experience
Lead SRE / Infrastructure Lead
Cybersecurity / highload infrastructure platform · 2024–present
- Lead infrastructure direction for a highload platform with carrier-scale L3/L4/L7 traffic and DDoS-mitigation workloads.
- Build and operate private cloud on Proxmox, Ceph, OpenTofu, Packer, Ansible and GitLab CI.
- Drive VM fleet migration from ESXi/vSAN-style to Proxmox/Ceph-style infrastructure with zero customer-visible downtime.
- Design multi-tenant observability with VictoriaMetrics, VictoriaLogs, ClickHouse, Vector and Grafana HA.
- Build BMP/BGP network telemetry, including SRv6, SR Policy, EVPN and FlowSpec.
- Own architecture, migration planning, cost models and operational standards.
- Provide technical leadership for a core team (2 direct reports) and set cross-team technical direction for ~5–7 network and adjacent infrastructure engineers by influence.
- Build and operate a private on-prem LLM/RAG inference platform — LiteLLM routing, vLLM + llama.cpp GPU serving, Qdrant + bge-m3 retrieval with reranking, Langfuse tracing — for SRE automation, incident response and MR review; adopted across multiple engineering teams, first external adoption via an IDE coding-assistant (Cline) integration.
Senior SRE / Core Infrastructure SWAT
Large-scale consumer-internet / highload platform · 2022–2024
- Core SRE/SWAT team handling critical infrastructure incidents and production degradations.
- Escalation point for P0/P1 issues across storage, compute and platform infrastructure.
- Stabilized Ceph clusters at ~100+ server and ~10 PB raw scale: disk failures, slow ops, recovery storms, rebalance, performance recovery.
- Migrated Ceph services from Kubernetes-style placement to bare-metal-style operation to reduce operational complexity and improve reliability.
- Contributed to Python services, CI/CD, PostgreSQL, Keycloak, Docker Swarm, object storage, performance tooling and observability.
Senior Cloud Infrastructure Engineer
International CDN / cloud provider · 2020–2022
- Operated and developed distributed IaaS on OpenStack, Ceph and S3-compatible storage.
- Supported 5 availability zones and ~80+ compute/storage servers; SLA for 50–80 B2B customers.
- Built resource accounting and billing pipelines for VM, storage, IP and S3 usage with PostgreSQL.
- Solved inconsistent S3 usage accounting with log-analysis pipelines.
- Ceph expansion, balancing, recovery, Ansible bootstrap, capacity planning, hardware selection and customer migrations.
Systems Engineer / Python Developer
GDS / travel-tech · 2018–2020
- Passenger-data processing pipelines replacing legacy shell workflows.
- SFTP ingestion, normalization, validation, retry and delivery control for external systems, with audit trail.
- Backend services in Python, Flask, Celery and PostgreSQL; query and index tuning.
- Asterisk/LDAP telephony integration and automation bots.
Projects
BGPeek — open-source BGP looking glass
- Multi-vendor SSH (Juniper JunOS, Cisco IOS/XE/XR, Arista EOS, Huawei VRP), RPKI validation, parallel cross-device queries with diff.
- RBAC (admin / NOC / public) with per-role output filtering; OIDC / LDAP / API-key auth; Fernet-encrypted SSH credentials; Redis rate limiting with circuit breaker; Prometheus /metrics; PostgreSQL audit log.
- Stack: FastAPI, Jinja2, HTMX, Tailwind, PostgreSQL (asyncpg), Redis, Netmiko, Docker Compose.
Private AI / SRE inference platform
- Operate a private LLM/RAG inference platform: LiteLLM routing, vLLM + llama.cpp GPU serving of Qwen-family models, Qdrant + bge-m3 retrieval with reranking, Langfuse tracing.
- Treat private inference as an infrastructure workload — NUMA-aware placement, hugepages, CPU/memory isolation, concurrency limits, context-budget sizing.
- Benchmarks for CPU-only isolation, GPU serving, prefix-cache behaviour and production co-tenancy impact.
Education
Moscow State Technical University of Civil Aviation — Information security of telecommunication systems