Kirill Shvakov

With over 20 years in tech, I design and architect high-performance distributed systems. I’ve built production systems in Go and led engineering teams—including AI agents as full participants in the process. I speak at conferences and have been at Kinescope for over 7 years, where I serve as CTO.

Articles & Talks

A 26-Minute Incident and 40 Cards in 12 Days: How Two Agents Became Part of the Team

How we brought autonomous agents into Kinescope's working chats: Ratchet in development and Hoist in infrastructure. Two paths of command execution (a shell inside the binary and argv without an interpreter), guards in code, agent code review instead of Bugbot, real incidents with timelines, an estimate of what this would cost in people, and the recipe.

Article

One Undefined Variable and 80 Minutes of CI: Why We Moved Our Ansible Rendering to Go

How a refactor of our Ansible variables ran into an 80-minute gating pipeline and ended with Go rendering: 85 specs across two environments in 3.5 seconds, byte-for-byte parity with Ansible, and a deployment plan on the same SDK.

Article

Speeding Up Go TLS to 100 Gbps: kTLS, Zero-Copy, and Production Gotchas

How to squeeze tens of gigabits out of a Go service over HTTPS: why TLS breaks zero-copy, how kTLS works, where moving encryption to the kernel helps, and what we had to change in production.

Talk

Why We Didn't Choose K8s: Staying Efficient and Improving Performance

Article based on a talk at TeamLead Conf: why we don't use Kubernetes in production and how we built operations on systemd, deb packages, and Ansible.

Talk

Operations Without K8s

Practical experience operating production infrastructure without Kubernetes: how we moved away from K8s and built a simple, reliable, and efficient system based on systemd and package managers.

Talk

Serving Content from HDDs: CDN Edge Evolution from File Chaos to Simple Storage

Practical experience building a CDN on regular HDDs: how to solve the problem of billions of small files, optimize disk operations, and how the system evolved from a complex architecture with merge to a simpler solution without background processes.

Talk

ClickHouse is Slow

Practical ClickHouse optimization experience: from performance bottlenecks to solutions using arrays and data structures. Scaling statistics system.

Talk

No Time to Explain, Just Code!

Practical development approaches and rapid prototyping techniques we use at Kinescope.

YouTube

How Go Brings Value to Business

A talk on the business benefits of Go: faster delivery, reliability, and performance.

YouTube