Software Engineer · MSc in Artificial Intelligence
I build production AI systems, open-source research tools, and developer-focused software — with a particular interest in multi-agent AI, LLM/VLM evaluation, retrieval systems, multimodal reasoning, and neural audio. I am based in Tenerife.
What I've Shipped
Institutional RAG system at Cabildo de Tenerife — one of Spain's largest public institutions:
- 🏆 Outperformed Microsoft Copilot Studio on document retrieval benchmarks against the institution's own corpus
- ⚡ Staff now surface institutional knowledge in seconds, not hours
- 🏗️ Sole AI Engineer — I owned architecture, retrieval pipeline, embedding strategy, LLM integration, and deployment end-to-end
▶ Case study: How I beat Microsoft Copilot Studio
The Cabildo already had access to Microsoft Copilot Studio — an enterprise product with significant backing. My task was to make their internal document corpus queryable by non-technical staff.
I built a custom RAG pipeline from scratch, making independent decisions on:
- Chunking strategy tuned to the structure of government reports
- Embedding model chosen for domain fit, not popularity
- Retrieval architecture designed around the query patterns of the institution's staff
In a controlled benchmark against the same corpus, my pipeline achieved higher precision and relevance than Copilot Studio. A government institution now runs on that system in production.
Research & Open Source
My current work combines practical AI engineering with independent research and open-source tooling:
- TrainLens — research-grade notebook reporting for AI training runs, directly inside Jupyter.
- MAVERICK — cognition-inspired multi-agent visual reasoning built around an auditable perceive → describe → critique → refine loop.
- Neural Audio Theory — an open educational project explaining the engineering foundations of AI music generation, from signal processing and embeddings to transformers, diffusion architectures, training, and prompt conditioning.
- music-to-text — local-first audio intelligence tooling that turns audio into structured metadata, A&R notes, PR pitches, playlist descriptions, and sync-licensing copy.
- LastLight — low-power, offline RAG for disaster guidance, designed around sourced retrieval and confidence-based refusal without cloud infrastructure.
- edujbarrios-ui — a frontend component library for clean AI interfaces, demos, and developer tools.
I also maintain a growing portfolio of research-oriented Python packages for model evaluation, embedding drift, VLM inspection, RAG chunk auditing, prompt safety, multimodal dataset quality, and text-to-music prompting.
Engineering Approach
| 🔍 Retrieval | Chunking and embeddings chosen per corpus — not per default |
| 🧪 Evaluation | Model and pipeline behavior measured explicitly, with reproducible diagnostics |
| 🤖 AI systems | Multi-agent and multimodal workflows designed for auditability, not just impressive outputs |
| ⚖️ Model selection | Cost / quality / latency tradeoff, not hype |
| 🏗️ Architecture | Observable, failure-tolerant, built to iterate |
| 🤝 Stakeholders | Requirements → specs → momentum across technical and non-technical teams |
Background
- 🎓 B.Eng. Computer Engineering (Software Engineering) — Universidad de La Laguna
- 📄 Thesis: 10/10 — Applied multimodal LLMs to retinal imaging for early glaucoma detection
- 🎓 M.Sc. Artificial Intelligence — Universidad Alfonso X el Sabio
Where I Work Best
✅ AI systems that need to move from experimentation into reliable real-world use
✅ Custom retrieval pipelines and RAG architectures built around the actual corpus
✅ LLM/VLM evaluation, multimodal reasoning, and uncertainty-aware workflows
✅ Multi-agent systems where roles, interfaces, and failure modes can be inspected
✅ Developer tools and research infrastructure that make AI experimentation more reproducible
Beyond Engineering
Electronic music producer EyeMad — 10M+ streams, featured on Spotify editorial playlists Electro Chill and New Music Friday. Building an audience from zero is a different kind of systems thinking: how creative decisions compound over time, and how to ship work that resonates.