Senior Solutions Architect
AI & Cloud Platforms

Senior Solutions Architect with 15+ years designing customer-facing systems in banking, telecom, insurance, and manufacturing - from discovery and stakeholder workshops through production, including North American delivery out of Toronto.
Recent proof: owned solution design for a company-wide AI PaaS that cut new-service time-to-market from ~3 months to ~3 weeks. Earlier: a 4-15k TPS SMS/voice anti-fraud platform for millions of subscribers, with availability-oriented design (multi-region failover and SLO/SLI) that sustained 99.95% in production, and Cloud Gate on AWS adopted by 70+ teams (release lead time ~6 months to ~3 weeks).
Hands-on in TypeScript, Java, Python, and .NET: I still write production code and PoCs, including a real-time SMS fraud model. I deliver C4, ADR/RFC, RACI, quality attributes, API contracts, and DR - and I coach teams to record whys.
Most effective when a platform or modernization has to be designed and shipped with engineering. Remote-first, aligned to US and Canada hours, English C1. AWS Certified Solutions Architect - Professional; also certified in AWS Security, Generative AI, Snowflake architecture, and Claude Architect Foundations.

Let's contact

I do work with

Main skills

Architecture & Design

Solution ArchitectureC4ADR/RFCDDDEvent-DrivenMicroservices

Cloud & Platform

AWS Well-ArchitectedKubernetesIAMAzureCI/CDObservabilityPlatform Engineering

Data & AI

AI PlatformRAG/GraphRAGLLM AgentsReal-time MLKafkaMCP

Engineering

TypeScriptReactAngularNode.jsJavaPython.NETRESTgRPC

Leadership & Delivery

Stakeholder ManagementPre-SalesMentoringGovernanceFinOps

September 2025 - Present

Sberbank - Largest Bank in Eastern Europe (100M+ clients) - Moscow, Russia

Senior Solutions Architect

Owned the target architecture for a company-wide AI PaaS so product teams consume RAG, agents, voice, and sandboxes as platform primitives - self-service for 17 teams, 8 services on the golden path, and MCP/REST gateways dropped new-service lead time from ~3 months to ~3 weeks.
To keep tenant blast radius and access scoped on the AI platform, applied AWS Well-Architected patterns - EKS-style cluster topology, IAM-bound tenancy, least-privilege access - and stood up CloudWatch-compatible metrics with Prometheus, Grafana, and OpenTelemetry so operations run against an SLO of 99.90% (SLI 99.95%), with MTTD of 2 minutes and MTTR of 20 minutes.
To modernize the AI model-components domain without a big-bang cutover, introduced DDD and an event-driven backbone (Kafka, Saga, CDC) and decomposed via Strangler Fig behind backward-compatible contracts so consumers kept working while the core was replaced incrementally.
Made architectural decisions reusable: ADR/RFC as the decision log - typically 1-3 ADRs every 1-2 weeks, review cycle 3-4 business days across the internal team, security, infrastructure, operations, and DevOps - and coaching so teams document why.
Led B2B/B2G discovery and POC/MVP cycles with C4 target-state, RACI, quality attributes, and delivery roadmaps.
Designed availability-oriented architecture for production - resilience patterns, multi-region failover, SLO/SLI - and validated behavior with Prometheus, Grafana, and OpenTelemetry.
To keep agents inside the bank perimeter, put MCP tools on an allowlist, ran untrusted code in secure sandboxes, and audited calls through API gateways - so teams used the platform.
Gated RAG and agent releases with an eval harness and LLM-as-judge, scoring changes on MRR, NDCG, precision, recall, and latency before they reached production.
Put FinOps controls on the self-service AI PaaS - team quotas, model routing, and shared primitives - so product teams did not open a separate GPU/LLM bill for every one-off stack.

April 2023 - September 2025

Beeline - Major Russian Telecom Operator - Moscow, Russia

Lead Solutions Architect

To protect a core SMS and voice service from fraud at peak load, architected and led delivery of a 4-15k TPS anti-fraud platform, modernizing the telecom path in front of millions of subscribers. The platform covered 99.5% of traffic; the remaining 0.5% could not be guaranteed because other operators owned part of the verification chain. Call records were written at ring time and scored after, so live legitimate callers were not blocked - only spoofed-CLI fraud was filtered.
Stayed an IC on the critical path: wrote and shipped the real-time SMS classification model the anti-fraud architecture depended on - ~500 ms latency, scoring ~15% of traffic after cheaper rule and regex filters, with precision around 95% and recall around 90%.
Led monolith-to-microservices decomposition (Strangler Fig) of a closed proprietary stack and a Radius/Diameter-to-gRPC (protobuf) migration on Kubernetes so the anti-fraud path no longer depended on a single protocol stack and could absorb new government-regulator requirements.
Designed event-driven core (Kafka, DDD, Saga) with multi-region DR so regional loss did not take the SMS/voice fraud plane down.
Presented migration options to the CTO and Principal Enterprise Architect and won approval for a staged on-prem decomposition with canary and A/B testing instead of a big-bang cutover.
Ran ADR/RFC design reviews and 24/7 SLO/SLI operations so the anti-fraud platform stayed inside agreed error budgets after each stage.
Treated SLO breaches as input to ADRs, so 24/7 operations on the anti-fraud plane produced architecture changes.

January 2011 - April 2023

EPAM Systems - Global IT Consulting - Gomel, Belarus & Toronto, Canada

Senior Software Engineer → Solutions Architect

The IC foundation first, then moved to Solutions Architect: 12 years of delivery across finance, insurance, manufacturing, and media, with North American client-facing work in Toronto.
Built the IC frontend foundation over 8+ years in TypeScript - React and Angular in production at EPAM (React continued after), plus Node.js for APIs, BFFs, SSR, and builds - across document workflow (Wolters Kluwer), legal (SwissRE), and e-commerce (COX Automotive), with overlapping teams of 10-50. Shipped microfrontends, SPAs, monorepos, and a plugin model for dynamic or custom component trees.
Proved that path on client UI, not only architecture: Cloud Gate UI at Wolters Kluwer; a 7-product SwissRE legal suite joined through microfrontends, including a Clearing Manager UI I rebuilt end-to-end so underwriters relied less on manual work; and at COX Automotive I built the credit-rating and auto-loan UI myself from bank-backed backend data, then led the team on a multi-tenant US vehicle sales and search portal.
To keep first paint and interaction usable, implemented SSR for the first page load and tuned smoothness and responsiveness against Core Web Vitals (LCP, CLS, INP) using Lighthouse and NPM metric runners - shrinking images, fonts, and JS, and keeping heavy work off the click path. The 50 ms budget for visual UI updates held as default: late content did not shift layout (CLS), the first viewport did not wait on the whole bundle (LCP), and clicks did not stall the main thread (INP).
Designed the test pyramid: Jest unit and integration tests with snapshots, Cypress E2E in Docker, and Playwright for generation, runs, and failure triage through agents on a local machine or in CI. On business-critical UI paths I added pixel-perfect visual checks so a layout break could not ship as a silent regression.
Alongside delivery, ran customer-facing pre-sales for EPAM clients - architecture workshops, RFI responses, and POC design - so technical discovery stayed honest and buildable.
Cut SwissRE Clearing Manager analyst cycle time from ~3 months to near-instant validated output by designing a 7-products insurance platform that kept sensitive workloads on-prem while using Azure where it paid off.
Through Wolters Kluwer Cloud Gate cut release lead time from ~6 months to ~3 weeks for 70+ teams in 10 departments by automating cross-product dependency testing on AWS.
To make Cloud Gate adoptable, built core components from scratch - EC2, EKS, and dynamic infrastructure to run tests, score relevance, and collect errors - established CI/CD and AWS deployment practices, and mentored engineers on behalf of the R&D department so the platform was implemented in production.
Treated code review and technical interviewing as part of delivery quality (Certified Technical Interviewer).
In parallel served as Resource Manager for a team of 10: ran 1-to-1s, worked on personal fit and adaptation to the project and the company, and resolved conflicts both between people and between a person and the project - compensation, duties, and engagement. Taught colleagues the STAR method so they could assess their own work by situation, task, action, and result. Mentored about 300 people over 5 years through seasonal specialized courses.