Skip to content

Pablo Díaz-Jorge

Software Engineer & AI Engineer
AI agents, MCP and frontend architecture

I design and build scalable software with a focus on architecture and quality, and I apply AI engineering to development: agents, MCP and measuring what they add.

Curriculum vitae

About

What started back in high school as a simple experiment has grown into my true vocation. Today I work as a software engineer and AI engineer: I design and build scalable systems with a focus on architecture, thoughtful design and sound engineering practices, and I apply artificial intelligence to improve the way software gets built.

I am currently working as a Frontend Engineer at Infoport Valencia, where I contribute to the development of digital solutions that optimize port logistics operations. My role involves designing accessible, scalable, and user-centered interfaces with Angular and TypeScript, supported by technologies such as SCSS, RxJS, and custom component libraries. I also integrate AI-assisted development tools into the workflow to accelerate development, refactoring, and code review tasks while maintaining high quality standards.

Between 2023 and 2025, prior to joining Infoport, I dedicated a period to applied research and independent development focused on the convergence between frontend architecture and AI engineering. During this phase I designed and prototyped four open-source solutions: Attention Matrix Explorer, an interactive visualization of transformer attention matrices with D3.js and Web Workers; Semantic Search Engine, a corporate semantic search engine with PostgreSQL, pgvector and HNSW indexing; Cognitive Agent Orchestrator, an agent orchestrator with the ReAct pattern and real-time SSE streaming; and Token Cost Analyzer, an LLM prompt cost and tokenization analyzer that runs entirely in the browser thanks to WebAssembly.

Since 2026 I have taken that work into AI agents and open source. I published prompt-forge, which now gives a team's AI coding agents shared memory reviewed through pull requests, and Candle AI, a technical analysis dashboard with a deterministic Bayesian confluence engine and an optional LLM narrative layer. I contribute to rea, an open-source project for reverse engineering with agents with more than 130,000 GitHub stars, where several of my contributions have been accepted, and in rea-lab I measure, with a preregistered study, what it adds to a coding agent. I have also published two Claude Code mods, rea-casebook and impact-radius, and complete applications such as gym-log. I am now expanding my profile into backend and databases.

Previously, I was part of NTT DATA as a Full Stack Developer intern, where I worked on the design and implementation of an employee management platform with Angular and Java. I also interned as a data analyst at Sothis, where I contributed to database analysis and design for compensation plans.

Outside of coding, you will often find me working out at the gym, hiking in the mountains, reading novels, or enjoying good company at a bar with friends.

Professional Experience

Sept 2025 - Present

Indefinite contract - Full time

Frontend Engineer - Infoport

I contribute to the development of digital solutions that optimize port logistics operations, building accessible, scalable, and user-centered interfaces with Angular and TypeScript, supported by technologies such as SCSS, RxJS, and custom component libraries. I integrate AI-assisted development tools into the workflow to accelerate development, refactoring, and code review tasks, maintaining high quality standards. I work in an agile environment, collaborating closely with analysts and backend developers to deliver robust, high-value solutions for the sector.

  • Angular
  • TypeScript
  • SCSS
  • RxJS
  • Signals
  • Kendo UI
  • Web Components
  • AI-Assisted Development
  • Scrum
  • Jira
  • Web accessibility

Jan 2026 - Present

Own projects and open-source contributions

AI Engineer (open source) - Open source

I build tools for AI agents and measure what they add, in my own projects and in open-source contributions. In rea (130,000+ stars) I have three merged PRs, which brought the real Ghidra verification on Windows back to green and moved the Claude Code and Gemini CLI checks to native Windows, plus an open PR in the MCP server's input validation (Node.js and TypeScript), design proposals and verification of other contributors' PRs with real Ghidra. In rea-lab I built a reproducible environment, 40 challenges with known answers and a TypeScript benchmark; its preregistered study of 70 Claude Code runs, graded with no LLM as judge, showed that rea did not change the solve rate, doubled time and cost, and that every cited evidence record was authentic. I have published two Claude Code mods, rea-casebook and impact-radius, with CI on Windows, macOS and Ubuntu, as well as prompt-forge and applications built end to end with agents, such as Candle AI, TestMaker and gym-log.

  • TypeScript
  • Node.js
  • Model Context Protocol
  • Claude Code (mods and skills)
  • Agent evaluation
  • Angular 22
  • Vitest
  • GitHub Actions
  • Open source
  • Preregistered studies
  • Technical writing

Jul 2023 - Sept 2025

Applied research and independent development

Independent Software Engineer (AI & Frontend R&D) - Open Source Projects

Applied research and independent development period focused on the convergence between frontend systems architecture and artificial intelligence engineering. I designed and prototyped four open-source solutions: Attention Matrix Explorer (2023), an interactive visualization of transformer attention matrices with D3.js and Web Workers; Semantic Search Engine (2024), a corporate semantic search engine with PostgreSQL, pgvector and HNSW indexing; Cognitive Agent Orchestrator (2024-2025), a cognitive agent orchestrator with the ReAct pattern and real-time streaming via SSE; and Token Cost Analyzer (2025), an LLM prompt cost and tokenization analyzer with tiktoken in WebAssembly. Stack: Angular (16 → 19), TypeScript, RxJS, Signals, D3.js, Spring Boot, PostgreSQL, pgvector, Node.js, Express, Web Workers, WebAssembly, SSE, Docker, Ollama.

  • Angular (16 → 19)
  • TypeScript
  • RxJS & Signals
  • D3.js
  • Spring Boot
  • PostgreSQL + pgvector
  • Node.js & Express
  • Web Workers & WASM
  • Server-Sent Events
  • Docker
  • Ollama & Qwen
  • Open Source
  • R&D
  • Technical Documentation

Mar 2022 - Jul 2022

Internship - Full time

Full Stack Developer - NTT DATA Europe & Latam

As a Full Stack Developer at NTT DATA Valencia, I contributed to the end-to-end development of a corporate personnel management solution. I took part in implementing REST services on the backend and developed an interactive frontend module focused on dynamic visualization of structured data, ensuring code maintainability under the team's quality standards.

  • Angular
  • TypeScript
  • Java
  • Spring Boot
  • Scrum

Mar 2018 - Jun 2018

Internship - Full time

Data Analyst - Sothis

Database analysis and design for the compensation plan of the UNE of Consulting and Information Systems (CSI) and the UNE of Automation and Control (AC).

  • MySQL
  • Power BI

Projects

rea-lab

Public lab for reverse engineering with AI agents on Windows, built on rea and Ghidra. A reproducible environment with pinned versions and hashes, 40 C challenges with known answers and a TypeScript benchmark that isolates every run and grades with no LLM as judge. Its first study, preregistered and with 70 Claude Code runs, showed that rea did not change the solve rate (28 of 30 with and without), doubled time and cost, and that all 80 cited evidence records were authentic. Its data has led to issues and PRs for rea.

  • TypeScript
  • Node.js
  • Model Context Protocol
  • Claude Code
  • Ghidra
  • Agent evaluation
  • Statistics (Wilson, McNemar)

rea-casebook

Claude Code mod that shows, live in a panel, the evidence rea gives the agent: the Ghidra wait, findings, open questions and the evidence timeline. Findings are recorded through a tool that rejects any evidence_id rea did not deliver, and the session exports to Markdown or to an artifact. TypeScript, 43 tests and CI on Windows, macOS and Ubuntu. Submitted to the Claude plugin directory.

  • TypeScript
  • Claude Code (mod)
  • Model Context Protocol
  • Vitest
  • GitHub Actions (3 OSes)

impact-radius

Claude Code mod that holds risky shell commands before they run (rm -rf, git reset --hard, force pushes, migrations, global installs, curl | sh), shows what they would change and waits for you to press Run or Cancel. Works with Bash and PowerShell on Windows, macOS and Linux, in English and Spanish, with AAA contrast and CI on three operating systems.

  • TypeScript
  • Claude Code (mod)
  • Bash and PowerShell
  • AAA accessibility
  • i18n EN/ES

Candle AI

Professional-grade candlestick charting and technical analysis dashboard that runs entirely in the browser: no accounts, no subscriptions, no data leaving your machine. Combines a deterministic Bayesian confluence engine (scoring 15+ candlestick and chart patterns across 7 signal hierarchies) with an optional AI narrative layer that explains the analysis in natural language. The LLM never overrides the quantitative engine; it narrates the results. Built with Angular 22 (Zoneless + Signals), Lightweight Charts v5, Web Workers for non-blocking computation, IndexedDB caching, Pine Script v5 export, and 160 Vitest + 5 Playwright E2E tests. Developed using prompt-forge agentic infrastructure across 9 epics as a solo developer.

  • Angular 22 (Zoneless + Signals)
  • Bayesian Confluence Engine
  • Web Workers (Indicators)
  • Lightweight Charts v5 (TradingView)
  • LLM Integration (Ollama/DeepSeek/OpenAI)
  • Pine Script v5 Export
  • Playwright E2E + Vitest (165 tests)

prompt-forge

Shared, reviewed memory for a team's AI coding agents. When one developer's agent trips over something in the repository, it records a lesson; when the same lesson shows up for two or more people, a curation step turns it into a short rule that every teammate's agent reads, whether that is Claude Code, GitHub Copilot, Codex, Cursor or Gemini CLI. Promotion goes through an ordinary pull request, so a person always reviews it, and follows a ladder: fix the code, then add a check, then write memory, and only then create a skill. Just Markdown and git: an AGENTS.md block, four Agent Skills and a knowledge/ directory, with Bash and PowerShell installers that merge without overwriting anything and CI on Linux, macOS and Windows. Its 1.0 version was the infrastructure Candle AI was built on.

  • Agent Skills (agentskills.io)
  • AGENTS.md
  • Shared team memory
  • Pull request review
  • Multi-agent (Claude Code, Copilot, Codex, Cursor, Gemini CLI)
  • Bash and PowerShell installers
  • CI on 3 OSes

gym-log

Mobile-first workout tracker for the gym floor: calculated progression targets for the next session (double progression autoregulated by RIR), volume and stagnation analysis, report export, and all data kept on your own device. Angular 22 with Signals and zoneless change detection, Tailwind CSS 4, Chart.js and Vitest.

  • Angular 22 (Zoneless + Signals)
  • TypeScript
  • Tailwind CSS 4
  • Chart.js
  • Vitest
  • Local-first

TestMaker

Secure exam platform with multi-user access and encrypted question banks (AES-256-CBC). Built with Angular 21 (Signals) and a Vercel serverless API. Features authentication, topic-based session configuration, advanced keyboard shortcuts, professional PDF export of results, ES/EN internationalization, and 18 Vitest unit tests. Built with AI assistance to solve a real need: studying with questions protected against unauthorized sharing.

  • Angular 21 (Signals)
  • AES-256-CBC
  • Tailwind CSS
  • Serverless API
  • PDF (jsPDF)
  • Vitest (18 tests)
  • i18n ES/EN

Token Cost Analyzer

Client-side LLM prompt cost and tokenization analyzer. Runs tiktoken WebAssembly inside a Web Worker for non-blocking tokenization. Visualizes token distribution with D3.js bar charts and provides cost estimates for 10 LLM models. Built with Angular 19.1.8 (zoneless change detection), TypeScript 5.7.3, and D3.js 7.9.0. Published as open source in 2026, preserving the original 2025 tech stack.

  • Angular 19 (Zoneless)
  • WebAssembly (tiktoken)
  • Web Workers
  • D3.js
  • TypeScript
  • RxJS & Signals

Cognitive Agent Orchestrator

Cognitive agent orchestrator implementing the ReAct (Reasoning + Acting) pattern with real-time streaming via Server-Sent Events. Features a deterministic state machine that reasons, invokes tools (calculator, weather, search), observes results, and streams every thought to an Angular 18 zoneless frontend. Dual LLM provider support: simulated mode and Ollama + Qwen 2.5 Coder 7B. Backend: Node.js/Express with TypeScript 5.5.4. Published as open source in 2026, preserving the original 2024-2025 stack.

  • Angular 18 (Zoneless)
  • TypeScript
  • Node.js & Express
  • Server-Sent Events
  • ReAct Pattern
  • Ollama & Qwen 2.5
  • Docker

Semantic Search Engine

Fullstack semantic search engine with vector storage in PostgreSQL via the pgvector extension. Implements cosine similarity search with HNSW indexing over 384-dimensional embeddings. Backend built with Java 21 and Spring Boot 3.2.6; frontend with Angular 17.3.12 featuring deferred result loading via @defer. Infrastructure orchestrated with Docker Compose. Published as open source in 2026, preserving the original 2024 stack.

  • Java 21 & Spring Boot 3.2
  • PostgreSQL + pgvector
  • Angular 17 (@defer)
  • TypeScript
  • HNSW Indexing
  • Docker Compose

Attention Matrix Explorer

Interactive visualization of transformer attention matrices (Query, Key, Value) inspired by the paper "Attention Is All You Need". Renders real-time heatmaps with D3.js (Viridis color scale, tooltips) and offloads matrix computations (softmax, transpose, multiplication) to a dedicated Web Worker. Built with Angular 16.2.12, Signals (Developer Preview), and D3.js 7.8.5. Published as open source in 2026, preserving the original 2023 stack.

  • Angular 16 (Signals Preview)
  • D3.js (Heatmaps)
  • Web Workers
  • Transformers (Q/K/V)
  • TypeScript

My Portfolio

Personal portfolio built with Angular 19.1.7 and Tailwind CSS 4.1, featuring SSR, i18n (ES/EN), dark/light mode, and lazy-loading for a fast, accessible experience. Built with Angular Universal and Express.js for Server-Side Rendering in production.

  • Angular 19.1.7
  • Angular Universal (SSR)
  • Tailwind CSS 4.1
  • ngx-translate
  • Express.js
  • PostCSS

Palaze (BSc capstone, legacy)

Native web component library built with Stencil.js. Lightweight, easy-to-implement components with fast loading. Installable via npm or CDN. Includes dedicated documentation. Created as a final degree project to dive deep into web development.

  • Stencil.js
  • Web Components
  • TypeScript
  • HTML & CSS
  • Figma