I build and own production AI systems end to end. Most of the code I write sits in private and client repositories under NDA, so what's public here is deliberately small.

At work I own the systems I care most about: a multi-agent platform that became the foundation other teams build on, and a GenAI platform that cut document processing time by a third on average. Alongside these I've built a range of automation pipelines, from simple online workers to multi-domain operations running on distributed Amazon EC2. None of it is public, and it won't be.

Early in my career, while completing my master's in Computer Engineering, I was selected twice to build AI decision-support systems end to end: one for civil-court mediation, one for Latin grapheme recognition in mediaeval manuscripts. I can't show the code, but both are published:

• "Explainable AI for Agile Mediation Propensity Assessment" IEEE, 2024
• "Location of Simple Graphemes in Mediaeval Manuscripts based on Mask R-CNN", IRCDL, 2021

What's public and mine is Exocortex: an MCP server I built as a persistent memory and reasoning layer for Claude. I wanted my interactions to be shaped around what I'm actually trying to achieve rather than generic assistance, and I wanted memory I could structure myself, as real files in real folders, which unlocks far more than a flat memory store. It remembers what matters from our talks, and it gets sharper the more it knows. It began as a fix for my own workflow, and I've used it every day since.