AI Automation Engineer · Enterprise Developer · MSc Artificial Intelligence

I've spent 20+ years building enterprise software across manufacturing, community sports, and healthcare — the kind of environments where reliability isn't optional and technical debt has real consequences. Over the last few years I've shifted my focus to AI systems, not because it's trendy, but because it's where the genuinely hard and interesting problems are right now.

I build things that replace real work. That means production pipelines, not prototypes — systems where an LLM is a component in a larger architecture, not the whole story.

I'm currently finishing an MSc in Artificial Intelligence at the University of Liverpool. My capstone, WorkflowSynth, investigates whether LLMs can automatically discover and synthesize enterprise AI workflows from natural language business requirements — a problem I care about because I've lived the manual version of it.

Most of my recent work sits at the intersection of document intelligence, RAG pipelines, and agentic automation. I use LangChain and LangGraph for orchestration, Pinecone for vector search, and the Claude and OpenAI APIs for the model layer. On the automation side, I run self-hosted n8n for workflow orchestration and integration work.

The through line across all of it: making AI systems that behave predictably under real operating conditions — correct outputs, graceful failure modes (where faliure is expected and planned for), and enough traceability that you can debug what went wrong.

WorkflowSynth — MSc capstone research on automated AI workflow discovery via LLM-guided programme synthesis. LangGraph-based synthesis engine with n8n prototype validation.

Soccer Verdun AI Support — Multi-tool AI agent for a Montreal soccer organization. Handles multilingual customer support (FR/EN) using Pinecone RAG, Gmail ingestion, and seasonal business logic. Human review stays in the loop for edge cases.

AlterEco Publishing Pipeline — End-to-end publishing automation for Alternativas Económicas magazine. PDF articles go in, structured Drupal content comes out. Reduced processing time from 30 minutes to 2-3 minutes per article using Claude Vision API for layout-aware extraction.

My enterprise years gave me something most AI engineers are still learning: what production actually means. I've debugged systems at 2am, inherited codebases with no documentation, and shipped things that had to work on the first try. That context shapes how I approach AI architecture — skeptically, with an eye on failure modes.

I'm based in Montreal. I work in English and Spanish and a bit of French.

MSc Artificial Intelligence, University of Liverpool (in progress)