Padraig O'Brien

• Owned a production X-ray microchip inspection system end-to-end, from automated ingestion to user-facing interface, reducing manual review effort by approximately 80%.
• Delivered yieldHUB's first production ML capability: large-scale anomaly detection used daily by engineering teams and supporting automotive quality workflows.
• Redesigned a production ingestion pipeline after customer-reported performance issues, improving throughput by approximately 16× and reducing compute costs by approximately 60%.
• Designed and prototyped a modular MCP-based agent architecture, defining reusable LLM abstractions aligned with existing APIs and platform constraints.
• Developed ML systems directly with enterprise customers, refining ambiguous requirements and challenging infeasible approaches; represented yieldHUB at industry events and delivered technical presentations on AI in manufacturing.
• Designed human-in-the-loop workflows with explicit failure handling and debugged time-critical production issues to maintain reliability in high-stakes manufacturing environments.
• Mentored interns and supported internal adoption of modern ML and GenAI practices.

• Built self-service GCP data-transfer tooling for non-technical users and automated reliable, repeatable backups with Airflow scheduled jobs.

Adaptive investigation agent separating LLM planning/interpretation from deterministic computation, with evidence-level provenance, contradiction handling, replayable investigations, typed termination, and evaluation over SEC EDGAR and tabular data.

Reproduced GPT-2 124M from scratch (3.0503 validation loss; 0.3043 HellaSwag); paired-seed ablations found optimiser choices dominated architecture (WSD: −0.1034 validation loss vs baseline), with 95.1% scaling efficiency across 8 GPUs.

Deployable RAG platform with multiple retrievers, persisted offline evaluation, regression comparison, query tracing, metrics, CI evaluation gates, and end-to-end testing.

Multimodal RAG using transcript/frame ingestion, HyDE retrieval, cross-encoder re-ranking, and cited timestamped visual evidence.

Current coursework in Probabilistic Machine Learning, Machine Learning Systems, and Accelerated Natural Language Processing; planned study includes Advanced NLP, Robotics, and Reinforcement Learning, followed by an MSc dissertation.

Foundation in statistical inference, machine learning, feature engineering, model evaluation, probabilistic modelling, and applied work in Python, SQL, and R.