blarrcher

Fintech platform: payments, crypto wallets, virtual cards, merchant integrations. $10M+ monthly volume. Team of 9 backend engineers; first year on the shared team, then dedicated owner of the authorization core, leading 2 engineers. Stack: Rust (tokio, axum, sqlx, SeaORM) · Python (anyio, FastAPI, aiohttp, Django), SQLAlchemy 2.0, Tortoise ORM · PostgreSQL, ClickHouse, Kafka, Redis (pub/sub, streams) · Kubernetes, Helm, GitLab CI/CD - Rust services in production since October 2025. Encryption service with exactly-once delivery (transactional Kafka producer, ordered offset commits) - replaced a slow and costly Python implementation. Notification gateway on a transactional outbox with consumer-side deduplication: legacy templates migrated in a day; new notification policies go live from the admin panel with no release. - Blockchain monitoring & custody automation (TRON). Rewrote a critical chain-monitoring daemon: a state machine tracking chain position with rollback after empty blocks (reorg-safe processing), adaptive backoff against node API limits, AML screening with automatic transfers to cold wallets. Transaction losses: from ~50/day down to zero. - Card authorization core. Designed and own the real-time approve/decline service (Python, anyio): balance debiting in a strictly consistent flow, 1-second hard SLA (fail-closed), 10,000+ RPS at peak. Removed the write ceiling (WAL contention on hot balance UPDATEs) by offloading read traffic to replicas and ClickHouse. Ownership backed by process: balance-invariant tests, alerting, ADRs, code review. - Zero-downtime migration from AWS to Hetzner. Planned and executed the move of stateful infrastructure (PostgreSQL, ClickHouse, Kafka, Redis, K8s) with no payment downtime: in-flight operations replayed through an orchestrator, topics synced via MirrorMaker2, DNS cut over through intermediate proxies without dropping incoming webhooks. - Payment provider integrations. Callbacks processed in a single database transaction: side effects (events, WebSocket) published only after commit, discarded on rollback. Idempotent operation handling under distributed locks. Hierarchical rate limits in Redis surfaced suspicious activity early and prevented a theft of funds. - Virtual cards. Shipped a production MVP in one month, replacing a legacy system and saving the product line; plugin-based architecture - 7 providers behind a single API, new provider onboarding in 1–3 days. Cross-currency operations with automatic FX-rate and fee calculation. Anti-fraud scoring. Real-time traffic scoring on Redis Streams; payment conversion grew from 43% to 71%. - Mentorship & engineering culture. Tech-debt backlog for tests, structured logging and metrics, code review on critical paths; conducted technical interviews; mentored three junior developers who now perform at a solid mid-level.

- Maintained and extended the CRM admin panel inside a large legacy monolith (Django + DRF). - Taught Python: 6 cohorts, 100+ students; still mentoring several graduates.

CRM for a government client (Russian Railways domain): normalized DB schema replacing a monolithic JSON-field table; event-driven interaction between 4 microservices. Accepted into production. Stack: Flask, PostgreSQL, SQLAlchemy, RabbitMQ, Redis, Docker, Jenkins.

Automated quarterly regulatory reporting: dependency-graph document parsing and a formula engine (~40 nested formulas per page) consolidating docx/Excel sources into 100+ page reports. Still in use by government agencies.

E-commerce store (Python, Django) and a business website (PHP).