Hi, I'm

Antonio Cabezas

I build the data infrastructure that keeps distributed systems fast and correct: Go, PostgreSQL, GCP/Cloud Run. For the past decade I've shipped products end to end, from solo builds to production systems.

Barcelona, Spain

role: Senior Backend Engineer

focus: Go · PostgreSQL · distributed systems

View my work Experience

Selected Work

Nimbus

Open-source · Go · Apache-2.0

Cloud Run-first caching for Go: an in-process L1, a versioned Redis L2, and a Pub/Sub bus that keeps caches coherent across ephemeral, autoscaling instances. It solves a gap no existing Go cache covered.

  • Correctness backbone: nothing enters the local cache without a version minted by the shared store, so concurrent invalidations cannot resurrect stale data.
  • Cross-instance invalidation over Pub/Sub (push by default, pull opt-in), designed for Cloud Run's scale-to-zero, CPU-throttled model.
  • Single-flight stampede protection, stale-while-revalidate, and tag-based invalidation, with benchmarks in the README.
Read the code Design doc

Lang Lab

Open-source · QEMU + GitHub Actions · MIT

Picking a backend language is a systems decision, not a habit. Lang Lab measures what the same algorithm costs across 13 backend languages by counting CPU instructions, not wall-clock time. The result is reproducible evidence for choosing the right tool per workload.

  • Differential instruction count: I(n₂)−I(n₁) subtracts process startup and JIT warmup, leaving only algorithmic work. The count is deterministic to the instruction, where wall-clock on shared CI runners drifts by 20% to 30%.
  • Correctness gate before any number counts: every implementation must reproduce a per-benchmark reference checksum bit-for-bit before its instruction count is accepted, so a fast-but-wrong solution can't reach the leaderboard.
  • 13 languages × 15 benchmarks, normalized to C = 1.00×: Rust 1.17×, C# 1.51×, Go 1.58× through Python 100× and COBOL 523×. On blur, Rust beats C at 0.48×, where LLVM auto-vectorizes and gcc -O2 doesn't.
Read the code Methodology

Background

I started in mobile in 2015, but kept gravitating to the other side of the API: the services, data, and infrastructure underneath. That is where I have spent the last several years, in Go, PostgreSQL, and GCP/Cloud Run, building systems where both correctness and latency matter.

Twice I have built products from zero to production solo, which is where I learned to make the tradeoffs between layers under real constraints. Today I work on backend systems and infrastructure at Forma Health.

Experience

Forma Health

Senior Backend Engineer

Backend services, data pipelines, and infrastructure on GCP for regulated digital-health products, with a focus on database performance and query optimization.

Forma Health homepage preview forma.health

Ciutadel

Founder & Engineer

Built a social network from zero to production, solo: backend, mobile clients, and data infrastructure on GCP. Took it to market, concluded it wouldn't hold as a consumer product, and wound it down.

Comloud

Founder & Engineer

My first solo product, shipped end to end: a marketplace to strengthen local commerce, built across backend and infrastructure. The economics never added up at the scale it needed, so I closed it and moved on.

Technical Stack

Languages

Go, SQL, TypeScript, Python

Data & Infra

PostgreSQL, BigQuery, Redis, GCP, Cloud Run, Pub/Sub, Kubernetes, Terraform, Docker

Domains

Healthtech, Social Networks, Logistics