Available for AI Automation & Engineering roles

Agustín Osilio

I build AI tools, end-to-end. agent pipelines with LLMs. things that ship to production.

AI Automation Engineer. I build tools and integrations end-to-end.
Designed, debugged, and shipped.
Backed by fintech and IT operations experience.
Based in Buenos Aires.

Scroll
01

Summary

Portrait of Agustín Osilio

I build tools and integrations that work, end-to-end, with AI as the main lever. When I hit a problem, I break it down until I understand which technologies can solve it, design the architecture, and work it through until it runs in production, including the debugging and iteration. And I don't come purely from the technical side: my solid experience in IT operations and fintech means I read the business behind the code, and I know where to focus to prevent problems before they happen.

AI Automation
Built end-to-end: agents, routing, integrations and generative AI workflows.
Fintech & IT Ops
ITIL, fintech back-office, SLA management, cross-team alignment across multiple areas.
End-to-EndBuilder
From problem breakdown and architecture to production, debugging and iteration included.
02

Experience

  1. AI Automation Engineer (Independent / Open Source)

    Independent Projects

    Current — Present
    • Built and open-sourced an API translation proxy (Python, FastAPI, httpx) that runs Claude Code on NVIDIA NIM: bidirectional Anthropic↔OpenAI format conversion with full SSE streaming, tool calling, image support, real token accounting, and automatic rate-limit retry logic, all in a single auditable ~720-line module.
    • Implemented LLM routing across local and hosted models (Ollama, NVIDIA NIM) for agent environments, integrating bidirectional Slack communication (n8n) and database connectivity (MCP/DBHub) for real-time conversational analysis.
    • Engineered a real-time gesture control system for Windows (Python, MediaPipe, Win32 API): a 4-thread design decoupling capture, inference, and rendering; an FSM with debounce and EMA smoothing; native overlay rendering with multi-monitor support. Open-sourced on GitHub.
    • Deployed SonarQube on Docker and integrated it with Claude Code through MCP for AI-assisted static code analysis; published a step-by-step setup guide covering both standalone and MCP-driven workflows.
    • Developed end-to-end AI bots and tools with Claude Code, integrating REST APIs, WebSockets, and RSS to automate data processing, file organization, and web research using advanced prompt engineering.
    • Built advanced generative AI workflows in ComfyUI: image and video generation, LoRA training, and custom-model pipelines using Flux, Wan 2.2, and Qwen models.
    Python FastAPI Claude Code MCP n8n NVIDIA NIM Ollama Docker ComfyUI

  2. Back Office Lead — Fraud & Chargeback Operations

    DolarApp (Fintech, now ‘ARQ’)

    —
    • Managed ~240 fraud and commercial-dispute chargebacks per month end to end, analyzing transaction-level data (EMV, NFC, 3-D Secure, tokenization, MDES) under KYC and data-privacy constraints, achieving an 80%+ win rate.
    • Performed data analysis and reporting with SQL and Metabase to drive operational decisions, working cross-functionally with risk, compliance, product, and engineering teams to resolve issues and optimize processes.
    • Supported transaction operations, customer onboarding, and payment troubleshooting, managing communication with external partners to ensure SLA adherence.
    SQL Metabase KYC 3-D Secure Chargebacks SLA

  3. Advanced App Engineering Associate

    Accenture

    —
    • Managed incidents under ITIL standards using ServiceNow: real-time monitoring, SLA tracking, and handoff preparation across global teams.
    • Developed operational reports and metrics with Power BI and PowerPoint to support decision-making for new and existing clients.
    • Took expanded ownership of the AS400 (IBM i) operation for the same global clients: system administration, process audits for risk identification, and internal control.
    • Maintained documentation control (runbooks, guides, operational files) and coordinated IT-operations tasks across teams.
    • Worked with AWS EC2 (FinOps), Microsoft Endpoint, and Intune for security and compliance management; supported onboarding and training of new team members.
    ITIL ServiceNow Power BI AS400 (IBM i) AWS EC2 Intune

  4. Network Operations Technician

    Accenture

    —
    • Monitored and operated AS400 (IBM i) infrastructure for two global clients: a leading automotive manufacturer (auto-parts inventory and backup processes) and a Fortune 100 heavy-equipment manufacturer (billing processes).
    • Resolved and escalated ~100 incidents per month under SLA targets, documenting and coordinating handoffs across global teams.
    AS400 (IBM i) Incident Management SLA Monitoring
03

Projects

Real-Time Gesture Control System

Control Windows with hand gestures in real time. Built in Python with MediaPipe and the Win32 API: a 4-thread design decouples capture, inference, and rendering, with a finite state machine, debounce, EMA smoothing, and multi-monitor support.

Python MediaPipe Win32 API Threading FSM
Code Web Demo

NVIDIA NIM Proxy for Claude Code

GitHub repository preview card for NVIDIA NIM Proxy for Claude Code

A minimal proxy that lets Claude Code run on free NVIDIA NIM models by translating between the Anthropic Messages and OpenAI Chat Completions APIs. Built in Python with FastAPI and httpx, it handles streaming, tool calling, image input, token accounting, and retries with exponential backoff — all transparent to Claude Code, which just sees a normal Anthropic endpoint.

Python FastAPI httpx Anthropic API NVIDIA NIM
Code

n8n · Claude Code Agent in Slack

GitHub repository preview card for Claude Code as a Conversational Agent within Slack

Agent environment with LLM routing between Anthropic and NVIDIA NIM, bidirectional Slack communication via n8n, and live database access through MCP/DBHub for real-time conversational analysis over operational data.

Claude Code NVIDIA NIM n8n MCP / DBHub Slack
Code

SonarQube + Docker + MCP — Setup Guide

GitHub repository preview card for SonarQube + Docker + MCP — Setup Guide

A practical guide to running SonarQube locally in Docker for automated code-quality analysis. Covers two workflows: the traditional SonarScanner CLI feeding the dashboard, and an MCP integration that wires Claude Code to SonarQube for real-time, in-terminal issue detection and AI-assisted fixes, with steps for both Windows (PowerShell) and Ubuntu (bash).

SonarQube Docker MCP Claude Code Code Quality
Code
04

Skills

  • Claude Code
  • Cursor
  • Ollama
  • LLM Prompt Engineering
  • ComfyUI
  • n8n
  • Custom Agent Dev
  • SQL
  • Python
  • HTML / CSS / JS
  • Powershell / Bash
  • REST APIs / Webhooks
  • WebSockets
  • AWS EC2
  • SonarQube
  • SonarScanner CLI
  • QA
  • FastAPI
  • MediaPipe
  • NumPy
  • OpenCV
  • pywin32
  • Docker
  • Node.js
  • PostgreSQL / Neon
  • WSL2 + Ubuntu
  • Slack API
  • NVIDIA NIM API
  • Win32 API
  • MCP (Model Context Protocol)
  • Intercom
  • ServiceNow
  • Jira
  • Citrix Workspace
  • Metabase
  • Slack
  • Retool
  • Spanish — Native
  • English — Advanced
05

Education

2017 — 2020

Degree in Multimedia and Interaction Design

UADE — Universidad Argentina de la Empresa

Focus on UX/UI, digital product design, and emerging interactive technologies.

2011 — 2016

High School Diploma

Social Science

Foundations in social sciences, communication, and critical thinking.