Sergey Ivanov

Data consolidation service for the Gazprom-Media holding: multi-source ingestion, automated processing, delivery to the external perimeter. Replaced a legacy solution — from inherited services to a unified platform. Cross-functional project with 5 adjacent teams (QA, frontend, systems analysis, business analysis, design). Leading a backend group of 4 engineers, code-manager mode.

Scope

• Domain and integration architecture: module and system boundaries, exchange contracts, top-level data model, NFRs — approved project architecture
• Reliable data processing: PostgreSQL with RW/RO replication; idempotency on Redis with atomic Lua scripts; transactional outbox — requeue of dropped entities
• Delivery: requirements shaping and sign-off with product and analysts, planning, prioritization, code review; estimation process (story points, planning poker), Jira automation and team status dashboards; clear ownership and role boundaries across the team
• Security and operations: access role model (RBAC/ABAC), system SLA, integration and perimeter approvals with InfoSec
• Release cycle: CI/CD, release management, documentation
• Hiring and growth: tech interviews, onboarding, 1-1s, individual development plans
• Consulting holding teams on python architecture and MVP launches

Impact

• 90–95% sprint accuracy after introducing Agile with story points; throughput up ~1.5× in six months — the practice spread to other holding teams
• Monitoring on Grafana + Alertmanager: incidents used to surface via support tickets — now proactive alerts with 5–10 min response
• CI/CD on GitLab + ArgoCD: feature toggles, auto-semver, migration to Trunk Based Development — feature rollback without redeploy
• TSG — project standards: code, zero-downtime migrations (blue-green), API contract evolution (backward compatibility), commit conventions
• Agentic development project-wide: an LLM harness, prompts and skills shared across the project; AI code reviewer on pydantic-ai with an extension layer — engineers ship faster, with less time on reviews and QA bug triage
• Agentic knowledge base (agents + indexed in-repo docs and README) — streamlined onboarding and support workflows
• Run trainings and master classes on agentic development; consult adjacent teams and directorates on AI adoption

OSINT analytics platform: collecting, processing and searching open-source data — 5k+ sources with auto-discovery, 100+ GB rolling 30-day history, ML enrichment. Team of 5; acting team lead: task coordination and architectural decisions.

Scope

• Data collection system design: parsers, parser manager, data bus
• Processing and enrichment services, including ML integration
• Search capabilities: full-text and faceted
• Database support and complex SQL
• Internal guidelines and documentation

Impact

• Designed and built the parser manager — system throughput up to 1M posts/day over the Kafka data bus
• Event-driven Telegram parsers refreshed prod indexes in ~2 seconds
• Full-text search on PostgreSQL TSV + ManticoreSearch with high relevance; designed and shipped faceted search — the key feature for analysts
• Rewrote heavy SQL via SQLAlchemy Core — DB load down 10×+
• Guidelines and docs became the foundation for team scaling

Backend services for client business automation at myradius.ru: from CRM integrations to accounting and monitoring systems.

Scope

• REST API development on FastAPI and Flask
• Async background tasks: Celery, RabbitMQ, Redis
• Database design and migrations
• Unit tests and code quality

Impact

• Migrated the stack from Flask to FastAPI — unlocked async and scalability
• Idempotent async workers — processing time down 40%
• Unit tests with >92% coverage — improved service stability

Full-cycle custom development: REST APIs, scrapers, business automation — from requirements analysis to deploy. 20+ projects delivered; direct communication and transparency with clients.

Scope

• REST APIs, parsers, automation systems — full cycle
• Integrations: Telegram, Google Sheets, email, CRM
• Anti-bot engineering: JavaScript challenges, authorization, captcha; reverse-engineered JS — hardcoded tokens, phone encryption via a character map

Impact

• Automated recurring client workflows (CRM, reporting, notifications) — routine time down 70–90%
• hh.ru vacancies and resumes scraper for an ML dataset: a continuously updated dataset of 15M records

Built on mihomo: a custom parser prepares the server list; mihomo health-checks all 350+ nodes and always routes through the fastest one.

• Parser turns provider configs into a mihomo-ready server list
• Health-check across 350+ servers — the fastest node is always active
• Routing rules split RU / non-RU traffic

Self-hosted services on a home NAS — 24 containers behind a reverse proxy (Nginx Proxy Manager), managed via Portainer.

• Media pipeline: *arr stack → Jellyfin (59 series, 2k+ episodes)
• Photo library on Immich: 75k photos / ~1 TB, 4 users
• AdGuard Home system-wide with encrypted upstreams (DoT/DoH): 65% of DNS queries blocked (46k of 70k)
• Home network L2/L3: segmentation and routing for services and clients
• Homepage dashboard aggregates health and stats of all services