I build infrastructure that must not fail — and stop outages before they happen.

Distributed Systems · Reliability · Scalability · Go · Rust · Ethereum

🌐 happygopher.nl · 📄 Résumé · ✍️ Writing · (also known as @screwyprof)

I build high-load, fault-tolerant backends in Go and Rust, and ship fixes upstream into the open-source infrastructure they run on. Every claim below is backed by a merged PR, a public case study, or a number you can check.

I make reliability systematic, not reactive. I've taken fragile financial, e-commerce, and blockchain backends and made them scale predictably and recover on their own — pairing deep engineering discipline with simple, legible design. I set technical direction as a hands-on IC: standards, RFCs/ADRs, and CI gates, not headcount.

Public and independently verifiable — issue numbers, merged PRs, measured impact.

Every gRPC call in the official Rust client (formerly EventStoreDB) paid a ~40 ms latency penalty, silently slowing every read, append, and subscription on low-latency networks. Root cause: the transport never enabled TCP_NODELAY, so Nagle's algorithm and TCP delayed ACK stalled each request behind a kernel timer. My one-line fix dropped the median read from 42 ms to 311 µs (~135×) and was merged upstream. Issue #232, PR #233, and the write-up.

Validator updates re-decrypted every keystore on every call — an O(n) scrypt password-hash that restarted validators every ~2 hours and blocked institutional upgrades. I made the path O(1) (skip decryption when there are no local keys); upgrades resumed at 99.9% attestation, unblocking tier-1 staking providers. Issue #4936, PR #4126, and the write-up.

Introduced a per-validator relay architecture — relays configurable per validator, hot-reloadable via an atomic lock-free swap, with the Prometheus instrumentation it originally lacked. ~26% lower latency. Issue #455, PR #470, and the write-up.

Patched go-ethereum to run a Tendermint-style BFT consensus layer with delegated PoS and 1-second blocks — a working prototype before Ethereum's Merge.

Full history, with metrics and context, lives on happygopher.nl/resume.

Built the multi-tenant staking backend behind MetaMask and institutional clients from the ground up. Correctness-first API: row-level-security isolation, idempotent writes, a checkpointed state machine, and end-to-end audit trails — no duplicate side-effects, safe resume after partial failures. Established SDLC practice as an IC — OpenTelemetry-by-default, RFC-first features, ADRs, CI gates — adopted across my team and picked up by adjacent ones. Deep dive: Building Financial Infrastructure That Must Not Fail.

Migrated the product-catalog domain from a legacy PHP monolith to distributed Go microservices — 16k+ RPS/instance, p95 <100 ms, 99.9% through 11.11 / 12.12 mega-sale surges across six markets. CQRS read/write split over sharded MySQL; advocated Clean Architecture and TDD across teams.

Rebuilt a fragile monolithic trading platform into an event-driven system with DDD, CQRS, and Event Sourcing — critical paths from minutes to milliseconds, deterministic verifiable records, FSB/FSTEK-audited, DDoS-hardened. Introduced CI/CD pipelines and structured on-call, setting the org's reliability standards.

Pioneered verifiable, unbiasable on-chain randomness — a threshold-BLS random beacon built into Tendermint consensus, delivered end-to-end (patched Tendermint, a Cosmos SDK app, and a Go DKG library).

Personal projects showcasing systematic approaches to reliability, architecture, and testing.

• payment — a complete DDD / CQRS / Clean Architecture implementation: event-sourced aggregates, invariant enforcement, full test coverage.
• cqrs — event-sourced CQRS library with a Given/When/Then DSL for business-readable domain specs — a rare pattern in Go.
• form3api — Form3 API client with GitHub-style HATEOAS pagination and executable documentation as acceptance tests.
• delegator — high-performance Tezos delegation service (CQRS): 15 s full sync of 760k+ delegations, <1 ms API responses, 92% coverage.
• favkit — Rust CLI for macOS Finder favorites, with ADR-based architectural reasoning and Nix reproducibility.

Full archive on happygopher.nl/writing.

• Making Operational Decisions Explainable
• Designing Hot-Reloadable Per-Tenant Routing
• A Password Hash on Every API Call: The Bug That Restarted Our Validators Every Two Hours
• What Actually Moved the Score: A Rust Auth Server on highload.fun
• TCP_NODELAY: One Line, 135x Faster in KurrentDB's Rust Client
• Why VS Code can't install extensions to a remote when your local ones are read-only
• Building Financial Infrastructure That Must Not Fail

Open to senior and staff backend & distributed-systems roles — reliability, architecture, and scale. Remote-only, hands-on IC, no people management.

I keep complex business domains safe to change and easy to reason about, and raise the engineering bar by proof, not mandate.