Portfolio

I build backend systems, AI products & internet experiments.

Hey, I'm Ramiro — backend engineer at Mercado Libre and a builder who ships. I turn ideas into real systems and products people actually use.

01

Selected builds.

01

Compass

Security layer for agent payments — validates, authorizes and caps transfers before they touch the real rail.

Builder · Agent Payments · 2nd place hackathon · Dev3Pack accelerator $ loading clone count...
In progress
02

Nana Wallet

Accessible agentic wallet built for the Aleph Hackathon. It turns everyday-language transfer requests into a preview, explicit confirmation and WDK execution, with recipient memory that keeps addresses out of search results.

Co-builder · Tether WDK · React · Capacitor · Fastify $ loading clone count...
Hackathon project
03

Khora

AI-native People Analytics and organizational intelligence. Goes beyond dashboards — predicts, signals and acts.

Founder · AI · HR Tech $ loading clone count...
Closed
04

Hackvault

Peer-to-peer secret vault with encrypted persistence and bidirectional .env sync. Teams connect directly through HyperDHT and keep offline encrypted replicas.

TypeScript · HyperDHT · Hypercore · AES-256-GCM $ loading clone count...
Open source
05

ESP32 Voice Agent

Push-to-talk voice agent built on an ESP32-S3. The firmware records and plays audio while a FastAPI bridge handles speech, persistent Hermes sessions and short-lived TTS files.

ESP-IDF · C · FastAPI · Proxmox · Embedded audio $ loading clone count...
Hardware prototype
02

Papers & writing.

Apart Research — Global South AI Safety Hackathon 2026

S-OWMI: A Perimeter Auditing Framework for Open-Weight Models in Latin American Institutions

We propose the Spanish Open-Weight Maturity Index (S-OWMI), an auditable perimeter-evaluation framework for open-weight LLMs deployed by Latin American institutions that need data sovereignty and regulatory compliance. We evaluate Llama-3.1-8B and Qwen2.5-7B on a Spanish-diverse dataset against an English baseline.

With Mauricio Genta, Federico Hörl and Nicolás Adrián Oroz

AI Safety · Open-weight LLMs · Multilingual evaluation · Global South
+13.3pp
degradation in bias-detection consistency when moving from English to Spanish — a gap an English-only audit wouldn't catch.
Notes & posts
Tutorials on X
system.status · live · @ram4-dev
03

What I'm building in open source.

Real GitHub activity, straight from the API. Cached hourly. No makeup — a dead week shows up dead.

$ loading profile...
$ ls -la ./repos | head -9 view all →
$ fetching git log...

How many people clone these projects?

GitHub reports full clones and unique cloners for each featured repository.

$ fetching clone traffic...

Rolling 14-day window · updated hourly by GitHub · not a lifetime total

Let's build something real.

Building something ambitious in AI, backend infra or early-stage product? Let's talk.

Say hi 👋