Not a portfolio — an evidence library. Every item states the problem, approach, result, and what evidence is available.
Architecture decision
Problem A B2B payments platform needed aggregation, analytics, and metrics over growing operational data; running analytical queries on the transactional PostgreSQL database was slow and put the payment path at risk.
Result A reporting system that serves analytics and metrics without touching the transactional path — cutting report generation from ~30s on PostgreSQL to under 2s on ClickHouse, and staying fast as volume grew past 1M+ events/day.
ClickHouseKafkaGoPostgreSQL
System
Problem A B2B payments platform needed new product capabilities — billing, email notifications, cryptocurrency integrations — without destabilizing the existing system or risking the money-critical payment path.
Result A fault-tolerant microservice architecture where new services ship without risking the core payment path — with zero double-processing of payment events under retries. Led the team of 4 engineers delivering it.
GoGingRPCPythonFastAPIKafkaRedisPostgreSQLDocker
System
Problem Visitors want a fast, honest answer on whether there's a fit — but generic chatbots hallucinate, leak their system prompt, and can be hijacked by instructions hidden inside user messages (prompt injection).
Result A live, self-hosted production LLM system — not a demo — serving real visitor traffic on this site, grounded enough to cite its sources and safe enough to refuse out-of-scope or injected requests.
GoAstroOpenRouterRAGSSEMCP
Architecture decision
Problem A growing e-commerce platform was built as a monolith; deployments were risky, scaling was all-or-nothing, and the system was hard to extend for international expansion.
Result A scalable, fault-tolerant, easily extensible system ready for international expansion — migrated live at ~100% uptime, with the business never pausing delivery.
PythonFastAPIPostgreSQLRedisDockerTraefik
System
Problem An e-commerce platform expanding internationally needed its catalog and content localized into 30+ languages; manual translation could not keep pace with content changes.
Result Continuous localization into 30+ languages without manual steps in the loop, processing content changes as they happened — an early production example of the supervised-automation pattern I now build with LLMs.
PythonCeleryYandex Cloud TranslatePostgreSQLRedis
System
Problem An agency platform shipped through manual deployments with no systematic error visibility — releases were slow and production issues were discovered by users.
Result Repeatable, automated releases and production errors surfaced to the team before users reported them — cutting time-to-detect from hours to minutes.
GitLab CIDockerDocker SwarmSentry
Plugin / Tooling
Problem AI coding tools give inconsistent results out of the box — quality drifts, context gets lost between sessions, and token costs grow with project size.
Result A daily production workflow where AI handles implementation mechanics under engineering controls — the practical base for the AI development culture service.
Claude Codeagent pluginshooksMCP