Oleg Mihajlov

• Key backend engineer building the anti-fraud rules platform from an early-stage prototype to production readiness, owning critical subsystems.
• Cut hot-path p99 latency by 37% through rule execution deduplication, compilation caching, and query parallelization — optimizations that also halved monthly database costs.
• Led the migration of analytics reads to ClickHouse — keeping fraud analysts' dashboards fast as data grew and shielding the real-time scoring database from heavy queries.
• Architected and led the migration to an in-app authentication and user management system, replacing the legacy auth flow, and mentored mid-level engineers joining the team.

• Worked on dashboard infrastructure and agents.

• Returning intern on the Observability team, building tooling on Stripe's company-wide metrics platform across the full stack — stream processing in Java/RxJava on Netflix Mantis, a React frontend, and Terraform infra over Prometheus and AWS.
• Built a self-service tool that lets on-call engineers pull detailed info for any metric on demand, surfacing which subsets of machines are degraded during an incident instead of digging through logs.
• Piloted it with 2 engineering teams against live incidents.

• First internship on Stripe's Observability team, building cost-attribution tooling for the company's Prometheus platform.
• Built a self-service tool that attributed each team's Prometheus metrics and exposed their cost in a UI, so teams could prune their own expensive, high-cardinality metrics instead of the observability team manually hunting them down and chasing each owner.
• Adopted across teams and driving recurring annual cloud savings — realized before the internship even ended.

• Improved the Podcast Episode page with data-driven design changes — 4.3% increase in user retention and 15% boost in engagement within 3 months.
• Cleaned up 4,000+ lines of obsolete code and brought test coverage above 90%.

A native iOS reader for MangaDex, focused on CoreData persistence and UI performance. Grew to 8,000+ downloads.

A Go service that decides on every page view whether the visitor is a bot; in production at a small online retailer. Rules are JSON data edited in a React admin UI — v1 had them hard-coded — and any draft rule can be backtested on 30 days of logged traffic before it goes live. Signals: JA4 TLS fingerprints, VPN and datacenter networks, rate counters, crawlers verified by reverse DNS, and fpscanner flags the site reports after page load. Built AI-first: I defined the behavior and the blocking policy, AI wrote the code.