Software Engineer · Distributed Systems & Machine Learning

I build backend and distributed systems — and I ship them.

Sixteen years of software and infrastructure engineering. I co-founded a technology company and led it as CTO, shipping an ERP product adopted by ~25 companies; today I run my own ERP in Go, used by four businesses and integrated with Paraguay's national e-invoicing system. Eight years of R&D on the compute, networking and data layers of a cloud platform. Now moving into machine learning — models trained from scratch, attention written by hand to match PyTorch, and an LLM assistant with guardrails and an evaluation bank inside a real product.

Asunción, Paraguay · Remote · UTC-3
Spanish (native) · English (full professional proficiency)

Get in touch Download CV (PDF)

Selected work

Three things I can show, explain and defend line by line.

Own product · in production

chipa — an ERP in Go

Invoicing, stock, customers and reporting for small businesses in Paraguay. Go and gRPC behind Envoy, CockroachDB, a Next.js dashboard. Four businesses; integrated with SIFEN, Paraguay's national e-invoicing system (signed XML, batch submission, credit notes), approved in production and operated by me alone. 1,000+ Go tests, ~1,000 API and end-to-end checks, CI. It ships an LLM assistant with ten tools, a control leash, a per-question audit ledger, hard spend caps and a 27-case evaluation bank run against three models.

GogRPCCockroachDBEnvoyNext.jsLLM toolingOpenTelemetry

chipaerp.com ↗

Client work · 2026

Investor-onboarding platform backend

For an investment firm serving European clients: KYC, document custody and GDPR data-subject rights. I wrote ~85% of the Go backend — 10 gRPC services, 72 RPCs, 430+ tests — on serverless GCP (Cloud Run, Cloud Run Jobs, Cloud SQL): crypto-shredding with a per-case KMS key, TOTP 2FA, four-eyes approvals and end-to-end OpenTelemetry tracing. Delivered in about four weeks.

GogRPCPostgreSQLGCPCloud KMSGDPRTOTP 2FA

Machine learning · hands-on

Models trained from scratch, and the checks around them

A CNN image classifier trained on a dataset I captured myself — dumb baseline measured first, time-based split so nothing leaks. Multi-head attention written by hand and verified identical to PyTorch's. A DQN implementation. The classifier served over FastAPI with a parity check against the training script. It's a short track record and I'd rather say so plainly; what I bring alongside it is the engineering.

PyTorchPythonComputer visionAttentionDeep Q-learningFastAPI

Experience

2026 – present

Independent Software Engineer — Backend, Cloud & ML Systems

Freelance and own products · Paraguay (remote)

Backend of an investor-onboarding platform on serverless GCP: ~85% of the Go backend, crypto-shredding with KMS, 2FA, four-eyes approvals, OpenTelemetry.
Author of chipa, an ERP in Go with four businesses, integrated with Paraguay's national e-invoicing system; LLM assistant with guardrails and an evaluation bank.
Machine learning work: vision classifier from scratch, attention by hand verified against PyTorch, DQN implementation, model serving with a parity check.

2018 – 2026

R&D Engineer — Cloud Infrastructure & Distributed Systems

Cortical.com Inc. · ipcore Datacenters S.L. · def.com — USA / Spain

Provider-side cloud R&D: the layer that creates virtual machines and their networks — someone has to build the system that creates those servers; that's the side I worked on.
High-performance systems layer with concurrency and multithreading (Go, C++) over distributed data engines (CockroachDB, TiDB, PostgreSQL); ported OVN's northd control-plane logic from C to Go.
QEMU/KVM virtualization with CPU/GPU passthrough, 40G overlay networking with hardware offloading (OVN / Open vSwitch, SR-IOV, NVIDIA BlueField-2 DPUs), Layer 3 network with BGP.
Benchmarked self-hosted LLM inference on in-house NVIDIA GPUs: memory bandwidth was the limiting factor — and knowing that was the point.

2016 – 2018

Co-founder · CTO · Senior Backend Developer

Tavatech S.A. · Paraguay

Co-founded a technology and IT consulting company and led its technical area.
Designed and commercialized an ERP product (Java / Spring) adopted by ~25 client companies; delivered projects for SMBs and large enterprises, including Unilever Paraguay.

2010 – 2016

Developer

RAN Networks S.L. · Spain

Software on Windows / Linux stacks in PHP, C# and C, taking on increasingly complex projects.

Education

BSc in Computer Science — Universidad Americana (in progress, graduation stage)
Computer Engineering — Facultad Politécnica, UNA (transferred credits)

Stack

Languages: Go · Python · C · C++ · Java · Bash · PHP
Concurrency & systems: Go concurrency (goroutines, channels) · multithreading · high performance · low-level C/C++
Machine learning: PyTorch · NumPy · convolutional vision models · attention (by hand) · deep Q-learning · model serving and evaluation (FastAPI)
LLM systems: tool-using assistants (guardrails, audit ledger, spend caps) · evaluation banks · self-hosted inference (llama.cpp, GGUF, hybrid GPU/RAM)
Cloud & infra: QEMU/KVM (CPU/GPU passthrough) · overlay networking · Docker / Compose · Kubernetes (deploying and operating workloads) · GCP (Cloud Run, Cloud SQL, KMS, Cloud Build) · advanced Linux
Data & RPC: CockroachDB · TiDB · PostgreSQL · SQLite · gRPC · protobuf · Envoy
Networking: Layer 2 / Layer 3 · BGP · OVN / Open vSwitch · SR-IOV · NVIDIA DPUs
Security: HashiCorp Vault / OpenBao · KMS envelope encryption · TOTP 2FA · TLS/PKI · GDPR data-subject rights
Observability: OpenTelemetry (traces, metrics, logs) · Jaeger

Contact

Available for remote contract and freelance work, and open to full-time remote roles with teams in Europe and the US. My timezone (UTC-3, Paraguay) overlaps well with European working hours.

Email · [email protected] LinkedIn GitHub