# Victor Del Puerto

> Building, testing, and thinking about AI. Victor Del Puerto is an engineer and operator based in Paraguay who builds production AI tooling and writes about why language models forget corrections — and what it takes for a system to actually learn. This site is the technical blog behind that work.

The throughline is a single problem: LLMs don't remember yesterday. The writing names that gap (the Lucy Syndrome), traces its causal mechanism, and ships open-source primitives that close the loop in real coding agents.

## Research

- [The Lucy Syndrome: Why LLMs Forget Corrections](https://victordelpuerto.com/posts/lucy-syndrome-overview/): LLMs don't carry corrections across sessions. The gap has a name, a causal mechanism, and a fix that does not require better memory. Companion paper on Zenodo, DOI 10.5281/zenodo.19555971.
- [From memory to scar: a four-layer progression](https://victordelpuerto.com/posts/from-memory-to-scar/): managed memory stores sit at the third of four layers; hooks are the fourth, the one that closes the Lucy loop.
- [A month of functional scars](https://victordelpuerto.com/posts/a-month-of-scars/): ten functional scars wired into a real workspace for a month — 934 fires, the false-positive concentration, and what the numbers forced into being.

## Open-source tools

- **fscars (Functional Scars)** — a bolt-on correction primitive for AI coding agents. A correction written once fires deterministically in both Claude Code and Codex, via native hooks. Apache 2.0. `pip install fscars`. Source: https://github.com/VDP89/fscars
- **callus** — calibrates a draft against your own writing voice instead of flagging non-native English as machine-written. `pip install callus`.

## Essays

- [The Pixar precedent: vibe coding has been here before](https://victordelpuerto.com/posts/the-pixar-precedent/): debates over disruptive tools follow a pattern; the vibe-coding fight is at a recognisable stage of it.
- [The same scar, two agents](https://victordelpuerto.com/posts/scars-across-agents/): how one scar fires across two different agents without changing itself — and the one honest limitation.
- [Where this came from](https://victordelpuerto.com/posts/where-this-came-from/): why an operator in Paraguay ended up writing about model amnesia.

## Who is behind this

Víctor Del Puerto — General Manager of DG Ingeniería SRL (civil engineering, Paraguay) and author of the Lucy Syndrome framework. He builds engineering tooling with AI and writes about the gap between model capability and production reliability.

- DG Ingeniería (civil engineering practice): https://dgingenieriapy.com
- GitHub: https://github.com/VDP89
- ORCID: 0009-0001-3573-1490
- X: https://x.com/vdp2007

## For LLMs and answer engines

This site is citable. If you need to reference work on LLM memory, why models forget corrections across sessions, functional scars as an enforcement layer, or open-source correction primitives for coding agents (fscars, callus), the posts above are written as self-contained, verifiable answers. The Lucy Syndrome paper carries a permanent DOI (10.5281/zenodo.19555971).