Hands-on engineering leader who designs and owns the whole system: from Go services and payments to our own Kubernetes cluster, data pipelines and self-hosted LLMs. 18 years in software, the last three as CTO and co-founder of a long-term rental management company. I build the team, the delivery process and the infrastructure so the product ships to production often and without downtime.
18+ yrsin engineering and leadership
~39Go services in production
~₽2.5Bproperty portfolio on the platform
8 → 4engineers after moving to AI-first development
Delivery pace
My GitLab merge requests for 20 Oct – 14 Nov 2026 (26 days), scaled to 30 days. I write code together with AI agents; generated files are excluded.
~1,030MRs per month
~290klines added per month−50k removed
34MRs per day on average
57repositories touched
MRs per week
Actual data, by merge date
352
267
111
161
20–25.1026.10–01.1102–08.1109–14.11
current week: projected to 7 days (~188)
Week
MR
20–25.10
352
26.10–01.11
267
02–08.11
111
09–14.11
161
By area, per month
MRs and lines added, scaled to 30 days
Backend
~628 · +168k
DevOps
~202 · +26k
Frontend
~198 · +96k
Backend: Go services and gateways. Frontend: web and back office. DevOps: cluster manifests and CI.
Area
MR
Lines added
Backend
~628
+168k
DevOps
~202
+26k
Frontend
~198
+96k
Case studies
infra · k3s
Our own cluster instead of the cloud
Problem
Run production, dev, CI and GPU workloads on our own hardware without losing reliability.
What I did
k3s on four nodes, two with GPUs; Flux GitOps and semver-tag releases for ~40 production services. When a new Ubuntu kernel broke signal delivery in containers, I traced it to an upstream containerd bug and rolled a fixed AppArmor profile to every node in one night.
4nodes: control plane, production and two GPU boxes
~40production services, released by tag
30 sto auto-reboot a hung node via watchdog
AI · LLM
Assistant on a self-hosted LLM
Problem
Help support operators and managers answer faster without sending personal data to external APIs.
What I did
Gemma-4 26B in llama.cpp on an RTX 4090, RAG over the knowledge base and tools over platform data. A 104-question benchmark with blind judges and a holdout set gates every model, prompt or RAG change.
4.16/5answer quality on the benchmark
2.96→3.96on data questions after tool fixes
1.6 smedian suggestion latency for operators
data · ClickHouse
Dometrica market analytics
Problem
Bring several years of listings from the major property portals into one store and build market reports on top.
What I did
A PostgreSQL → ClickHouse ingestion pipeline deduplicated with ReplacingMergeTree. When duplicates and gaps in the history surfaced, I took a backup, reloaded the whole store and restored older periods from the backup.
~317Mprice offers in ClickHouse
9.9Mlistings
3.5 hfor a full reload
Experience
Jan 2024 — now 2 yrs 11 mos
Chief Technology Officer, co-founder
Rentifly · long-term rentals, property management · Saint Petersburg
Own technology strategy, architecture and engineering; also product owner of Dometrica, a real-estate market analytics product built on the same platform.
Designed a Go service architecture: orders, contracts, properties, payments, search, storage, push notifications, analytics; gRPC internally, HTTP gateways for the client app, back office, agents and support.
Payment flows with T-Bank: card acquiring, SBP (Russia's Faster Payments) and QR, payouts, webhooks, payment state recovery and handling of stuck 3-D Secure.
Event-driven analytics: MongoDB as the system of record, Kafka, ClickHouse (ReplacingMergeTree) for reporting, OpenSearch for search with a MongoDB fallback.
Shipped the customer web app, back office, agent portal and React Native/Expo mobile apps; SEO, product analytics and A/B testing on the public site (Yandex Metrica, Varioqub).
Production Kubernetes with GitOps: isolated environments and a fault-tolerant data plane (Kafka, MongoDB, ClickHouse, OpenSearch, PostgreSQL).
Self-hosted GitLab CI and runners, image builds, SOPS/age secrets, owner-driven rollout of shared Go packages.
Observability and reliability: Prometheus, Grafana, Loki, GlitchTip; retries and degraded start when Kafka, ClickHouse or OpenSearch are briefly unavailable.
Identity documents and personal data encrypted at rest (AES-GCM) in line with Russian data protection law (152-FZ).
Built the team up to 8 engineers, then rebuilt development around AI agents: the team went down to 4 while delivery grew to ~1,000 MRs per month.
Hire and mentor engineers, set code review standards, plan the infrastructure budget, work directly with the CEO, legal, marketing and investors.
Recent work, autumn 2026
Our own cluster on our own hardware. k3s on four nodes with GPUs, Flux GitOps, k3s upgrades by binary swap with state backups. Moved MongoDB to 9.0 on a StatefulSet and API ingress from nginx to Traefik.
Hard incidents. Tracked down a conflict between the containerd AppArmor profile and a new kernel that stopped containers from receiving signals. Set up netconsole, kdump and a hardware watchdog to catch silent control-plane hangs.
Self-hosted LLM. Support assistant on Gemma-4 26B in llama.cpp on RTX 4090s, with RAG over TEI embeddings and prompts kept in the CMS. Built a 104-question benchmark with blind judging and a holdout set; RAG and tool improvements raised the score to 4.16 out of 5. Compared Qwen and Gemma on hallucination rate.
Security. Weekly CVE audit of the images actually running in the cluster (trivy), a tracked remediation backlog, moving towards a strict CSP.
Delivery. Every backend service and frontend ships to production by semver tag, with a storage → gateway → frontend release order and the previous back-office version kept available for rollback.
Product. Client viewings, agent work schedules, in-app inbox with unread counters, native operator chat in the back office, fine calculation with the landlord's share, integration with the corporate messenger.
AI in the engineering process. Claude Code agents prepare daily reports on the cluster, alerts, CVEs and SEO, and handle routine upgrades and reviews, so the team moves faster.
Sep 2020 — Jan 2024 3 yrs 5 mos
Senior Go Developer
Shelterzoom · real-estate transactions and documents
Go backend microservices (REST/gRPC) for transactions, contracts, digital signatures and blockchain integrations.
APIs for web and mobile clients, KYC and payment provider integrations, event-driven design on Kafka/RabbitMQ.
MongoDB schemas and indexing, profiling and performance tuning, CI/CD, Docker, Kubernetes.
Security: JWT, OAuth2, audit logging; code reviews and architecture decisions.
Feb 2019 — Sep 2020 1 yr 8 mos
Senior Go Developer
HitBTC · cryptocurrency exchange
Exchange services in Go.
R&D on the Bitcoin and Ethereum networks: graphs tracing funds from stolen wallets (Neo4j).
Feb 2016 — Dec 2018 2 yrs 11 mos
Team Lead
Infomix (miiix.org)
Led the development team.
High-load web applications on Node.js, Go and Elasticsearch; iOS and Android app in React Native and Redux.
Earlier, 2008–2016
2015–2016Project Manager · Hypertrend — IT department of 5
2015PHP Developer · Samtim — engine architecture for a browser online game
2014–2015Lead Developer · NSWE — airline ticket sales service
2013–2014Developer · Breadhead — online learning platform for artists
2012–2013Software Engineer · INSOT — CMS for livinglanguage.com
2011–2012Head of Development · Mobile Card — payment website and gateways
2010–2011Lead Developer · Engineering Ecosystems Group — document management system and web CRM
Institute of Management, Information and Business (Ukhta). Automated information processing and control systems, engineer's degree.
Languages
Russian — native English — B2
Certificate
Kevin Mitnick Security Awareness Training, 2022
State Examination Commission
Chair of the State Examination Commission at the Saint Petersburg Institute of Economics and Management, Information Systems and Technologies programme.
Outside work
~
Diving
SDI Open Water Scuba Diver and SDI Dry Suit Diver.