AI Engineer · Machine Learning Engineer
_{Dublin, Ireland}

AI engineer based in Dublin. I work mostly in Python, across LLM and agent systems, computer vision and multimodal deep learning.

I like building things end to end: the model and the data work, the API around it, the deployment, the tests and the monitoring. Most of what I have built runs in production for real customers, which is where I learned to care about latency, cost and what happens when something fails.

M.Sc. in Artificial Intelligence, 9.11/10.

Self-hosted, multi-tenant platform that answers the phone and replies on WhatsApp for real businesses — a hair salon, a mechanic's workshop and others. It has handled thousands of calls over recent months, with paying clients depending on it.

LiveKit Agents over SIP telephony with automatic cross-vendor failover on speech-to-text, LLM and speech synthesis; the official WhatsApp Cloud API with Chatwoot and hand-off to a human mid-thread; a multi-client booking API over Google Calendar; and a dashboard the business owner runs on a tablet. 890+ automated tests, CI, cost reconciliation and GDPR retention, on Docker Compose.

Python LiveKit FastAPI SIP WhatsApp Cloud API Cerebras Vertex AI Google Calendar Docker React

Personalised illustrated children's books, generated and then physically printed and delivered. LLM storytelling with content-safety filtering, image generation, automated cover and PDF assembly, and print-on-demand fulfilment.

Next.js Supabase Stripe Replicate

Location intelligence scoring commercial viability across 1,591 geospatial nodes in 48 neighbourhoods, fusing 28 data sources — hourly water telemetry, cadastral records, OpenStreetMap mobility, census, rents and tourism.

LSTM temporal models, a GraphSAGE graph neural network over the urban network, and HDBSCAN clustering, with SHAP explainability. Scores returned in under ten seconds.

PyTorch GNN LSTM HDBSCAN SHAP

Deep learning for fine-grained animal behaviour recognition in the wild.

Exploring Temporal Action Segmentation Techniques for Enhanced Bird Behavior Recognition · published — SOCO 2025, Springer CCIS vol. 2806. Frame-level and segment-level pipelines benchmarked over I3D, R(2+1)D, MViT-B and PDAN backbones. Code.

ViP-BiRd: Multimodal Fine-Grained Bird Behavior Recognition via RGB Video and Pose Data · under review — multimodal architecture fusing V-JEPA video representations with ST-GCN++ skeletal motion over a custom 10-node avian graph. Reaches 55.91% on Visual WetlandBirds against 49.46% RGB-only and 47.31% pose-only, showing the gain comes from the fusion itself. Master's thesis, graded 10/10.

Predicting Avian Occurrence in Mediterranean Wetlands · under review — second author. ValWet-Birds, a harmonised dataset unifying professional censuses, eBird citizen science and Raspberry Pi acoustic sensors across three protected wetlands (2010–2025, 395 species). LSTM occurrence models cut test MSE by 42.7% over census-only training.

M.Sc. Artificial Intelligence — University of Alicante, 2025–2026. Overall 9.11/10, taught entirely in English. Thesis graded 10/10.

B.Sc. Computer Engineering (Ingeniería Informática) — University of Alicante, 2021–2025. Computing specialisation: theory of computation, automated reasoning, computer vision and robotics, language processing. Final-year project graded 10/10.

Northern Arizona University, USA — Erasmus+ exchange semester, Fall 2024. Upper-division computer science in English: Artificial Intelligence, Machine Learning, Algorithms, Automata Theory, Virtual Worlds. Top 5% of the senior Algorithms class.

cpp-search-engine — a search engine written from scratch in C++ with no external libraries. Hand-written tokenizer state machine, inverted index with positional postings and disk spilling, Porter stemming, and BM25 / DFR ranking with TREC-format evaluation.

art-with-drones — a distributed drone-show system: independent processes coordinated over Kafka, sockets and an HTTPS REST registry, with failure detection, engine recovery and hybrid RSA/AES encryption.

Most of what I work on lives in private repositories — they are live commercial products with client data in them. The projects above are the ones I can open up, and I am happy to walk through the architecture of any of the others.

_{EU citizen · based in Dublin · open to AI/ML engineering roles}