I build AI systems and study how they behave. My work connects LLM evaluation and reliability, graph neural networks for reasoning, and enterprise AI architecture.
I run Research and Implementation (Resai), combining hands-on engineering with research and advisory work. Previously: Partner at Implement Consulting Group, Director at Accenture, and CTO and co-founder of Arcus Global.
Research & publications · Professional background & contact
- Research with a competitive result: my AFGCN solver won four of six approximate-track categories at ICCMA 2021: Preferred, Semi-Stable, Stage and Stable. Official results.
- From research to practice: 25+ years in enterprise technology, including co-founding Arcus Global and helping grow it from two founders to 100+ professionals.
- Writing and teaching: author and co-author of five book titles on Salesforce, AI and analytics; taught Generative AI at DIS Copenhagen.
Open a quick start to run a CPU model, build an analysis report or verify a recorded result.
| Project | What to inspect | Try it |
|---|---|---|
| AFGraphLib | Graph-learning components, framework data and a complete GCN training example. | CPU example |
| AFGCN | Grounded reasoning, structural features and pretrained graph-network decision solvers. | Run a query |
| ExplainableArgGCN | Attack/defence message passing, ranking objectives, baseline implementations and ablations. | Model smoke test |
| SycBench | Installable, streaming dataset transformation and validated experiment-log scoring. | CLI example |
| LLM Linguistic Steering | Adjective-effect estimation, explicit answer parsing and report generation; accompanying TMLR research. | Local report |
| AI Adoption & Absorptive Capacity | Two-survey statistical analysis with verifiable tables and pinned data provenance. | Verify a result |
- LLM behaviour: sycophancy, specification gaming, evaluation methods and agent reliability.
- Learning to reason: graph neural networks, abstract argumentation and efficient inference.
- Applied AI: agent architectures, retrieval, MCP tooling and enterprise integration.
More to explore: FastAFGCN · Argumentation subsampling · Multiple-Choice Question Generator
- Investigating Linguistic Steering: An Analysis of Adjectival Effects Across Large Language Model Architectures — Transactions on Machine Learning Research, 2026. Code.
- Sycophancy in Large Language Models: Causes and Mitigations — Computing Conference, 2025. Open preprint.
- Winning at All Cost: A Small Environment for Eliciting Specification Gaming Behaviors in Large Language Models — ACNS Workshops, 2025; link to open preprint.
- Approximating Problems in Abstract Argumentation with Graph Convolutional Networks — Artificial Intelligence, 2024, with T. Yuan and P. Nightingale. Code.
Full publication list and citation record on Google Scholar
Books and practitioner writing
- Practical Salesforce Agentforce Playbook — Packt, 2026.
- Salesforce Anti-Patterns — Packt, second edition, 2025.
- Salesforce DevOps for Architects — with Rob Cowell; Packt, 2024.
- Tableau for Salesforce — with Damiana Spadafora; BPB Publications, 2024.
- Architecting AI Solutions on Salesforce — Packt, 2021.
More practitioner writing: Salesforce Ben · Apex Hours.
PhD in Computer Science, University of York · MBA, University of Cambridge (Director's List).
Certifications: Salesforce Certified Technical Architect · TOGAF 9 · Claude Certified Architect (Foundation).
Based in Madrid, Spain (CET/CEST). Available for 100% remote AI architecture, research engineering, evaluation and advisory engagements, and research collaboration.