I design and build distributed backends — and I run them in production.
Platforms, Kubernetes operators and infrastructure in Go. What I care about is taking the human out of operations where being wrong is expensive — an upgrade that has to land at an exact coordination point set by something outside the cluster, a signer that refuses rather than risk a conflicting signature, storage that resizes itself. Systems with state, hard correctness constraints, and nobody watching at 3am.
Public, MIT, under @voluzi. These target Cosmos-SDK blockchain networks — that is where I have run production infrastructure for the past few years — but the engineering is not domain-specific.
-
Cosmopilot — Kubernetes operator for a stateful, network-coordinated workload. Watches an external source of truth for upgrade events, swaps the container image and restarts at an exact coordination point, bootstraps new replicas from remote state, verifies snapshot integrity, resizes PVCs in place, and integrates with a remote signing service.
-
CosmoGuard — API gateway. Endpoint-level access control over HTTP, JSON-RPC, WebSocket and gRPC; response caching; rate limiting per IP, globally, or per authenticated identity; API-key, JWT and RFC 7662 introspection auth; multi-upstream with active healthchecks and circuit breakers. Single binary — cache and rate-limit state are shared across replicas without an external store.
-
Cosmosigner — high-availability signing service for keys held in HashiCorp Vault or Cloud KMS. Every signature passes a raft-committed high-water-mark before it exists, so a replica that loses quorum cannot sign at all: it fails closed rather than risk producing two conflicting signatures for the same decision.
Go, Kubernetes, distributed systems. Based in Portugal.