TSI alumnus (IT, Class of '05). I build and run high-availability, high-concurrency edge infrastructure — zero-allocation Go network daemons, Linux kernel tuning (XDP, nftables), and aggressive codebase stripping to cut runtime overhead and attack surface.

My main project is DNSDOH.ART — a public encrypted-DNS resolver I build and run as a hobby. No logs, no telemetry, no company behind it. → ozy-666.github.io

The goal behind all of it: a single node that simply does not fall over.

• L4, kernel-space (eBPF/XDP): packet-filtering engines in Go that drop volumetric garbage (SYN floods, UDP amplification) at the NIC driver level — before the kernel ever allocates an sk_buff.
• L7, user-space (Go / nginx): zero-allocation byte parsers that rate-limit and sanitize malicious DoH / HTTP requests.
• Dynamic nftables orchestration: a Go control plane talking to nftables over Netlink sockets, pushing real-time blocklists straight into kernel-space sets to isolate abusers with minimal CPU cost.

A single, carefully-tuned server — not a global anycast network — running encrypted DNS over DoH, DoH3 (QUIC), DoQ and DoT, with ad/tracker blocking and a strict no-logs policy. The focus is reliability and zero tolerance for DDoS and abuse, not out-scaling the big providers.

• AdGuardHome-edge-spec — the engine: a stripped-down AdGuard Home fork (~13k LOC removed, zero-allocation hot paths, Unbound on BoringSSL).
• dnsproxy — transport fork: pooled connections, SO_REUSEPORT listener sharding, lock-free upstream RTT map.
• dnscrypt-proxy — encrypted-upstream fork: sync.Pool packet buffers (0 B/op on hot paths), monitoring compiled out, security-audited.
• urlfilter — filtering engine: AST-based required-literal extraction (O(1) regex miss paths).
• dns-ultra — a DNSCrypt / DoH benchmarking and auto-tuning suite.

Profile-driven: every change is proven with pprof / benchstat on real hardware before it ships, and the dead-ends get documented alongside the wins. Static analysis, profiling and refactoring done with AI tooling (Claude Code, Gemini / AI Studio CLI) in the loop.

QA & uptime inspection: Maine Coon Michelle 🐾 — strict about stability and allocation budgets.