---
title: "Ignacio Di Rocco (@idirocco) | Hashnode"
description: "Building products, exploring AI, and sharing the engineering behind them. Read the latest articles by Ignacio Di Rocco on Hashnode."
url: "https://hashnode.com/@idirocco"
---

@idirocco

# Ignacio Di Rocco

@idirocco · Argentina · Joined October 2026

Building products, exploring AI, and sharing the engineering behind them.

### [Why Generic Synthetic Examples Fail in Category-Specific Scenarios](https://idirocco.hashnode.dev/why-generic-synthetic-examples-fail-in-category-specific-scenarios?utm_source=hashnode&utm_medium=feed)

A targeted generator-prompt change, what 93.6% actually means, and lessons from AI Waiter

4 min read · [idirocco.hashnode.dev](https://idirocco.hashnode.dev/why-generic-synthetic-examples-fail-in-category-specific-scenarios?utm_source=hashnode&utm_medium=feed)

[6d ago](https://hashnode.com/posts/why-generic-synthetic-examples-fail-in-category-specific-scenarios/6aea14df5ea170c1ad922669)

### [A Small Pipeline, With Explicit Boundaries: Architecture & Implementation](https://idirocco.hashnode.dev/a-small-pipeline-with-explicit-boundaries-architecture-implementation?utm_source=hashnode&utm_medium=feed)

Model choices, orchestration, structured outputs, evaluation, and the production trade-offs I haven't measured yet.

6 min read · [idirocco.hashnode.dev](https://idirocco.hashnode.dev/a-small-pipeline-with-explicit-boundaries-architecture-implementation?utm_source=hashnode&utm_medium=feed)

[Oct 28](https://hashnode.com/posts/a-small-pipeline-with-explicit-boundaries-architecture-implementation/6ae1d3e7a82de6eb4e6a2671)

### [A Passing Eval Doesn't Mean Your AI Feature Is Reliable](https://idirocco.hashnode.dev/a-passing-eval-doesn-t-mean-your-ai-feature-is-reliable?utm_source=hashnode&utm_medium=feed)

How I think about evaluation sets, failure analysis, and the gap between a good score and a dependable AI product.

8 min read · [idirocco.hashnode.dev](https://idirocco.hashnode.dev/a-passing-eval-doesn-t-mean-your-ai-feature-is-reliable?utm_source=hashnode&utm_medium=feed)

[Oct 19](https://hashnode.com/posts/a-passing-eval-doesn-t-mean-your-ai-feature-is-reliable/6ad65d7184992924134251fd)

### [My LLM Judge Went From 70% to 96% Agreement With Humans. Here’s What I Learned.](https://idirocco.hashnode.dev/my-llm-judge-went-from-70-to-96-agreement-with-humans-here-s-what-i-learned?utm_source=hashnode&utm_medium=feed)

What three iterations of an LLM-as-a-judge rubric taught me about evaluation design, regression testing, and synthetic data quality.

14 min read · [idirocco.hashnode.dev](https://idirocco.hashnode.dev/my-llm-judge-went-from-70-to-96-agreement-with-humans-here-s-what-i-learned?utm_source=hashnode&utm_medium=feed)

[Oct 15](https://hashnode.com/posts/my-llm-judge-went-from-70-to-96-agreement-with-humans-here-s-what-i-learned/6ad0d9d3385c982bf108df04)