I lead the strategy, design, and implementation of AI systems and programs that drive efficiency and business impact.
With 11+ years across research, development and innovation in cloud-edge platforms and AI, I lead large-scale technological initiatives that maximize ROI, driving business success while reducing cost and environmental impact. My expertise uniquely bridges AI/ML systems, business strategy, and sustainability, the key combination needed to leverage AI effectively in modern organizations.
Highlights:
- Lead strategy & innovation initiatives for Telefónica O2 Germany’s Tech Directorate, focusing on edge, cloud, and AI
- Co-founder of Leaner AI, helping SMEs cut the cost, energy, and emissions of AI inference
- Recognized expert in AI energy efficiency: doctoral research at Universidad Politécnica de Madrid, 300+ citations, and several publications in journals and top conferences (VLDB, AAAI symposia)
- 9 patents filed (US/EU/Japan) in ML & systems optimization, 5 already granted
- Led funding proposals, TCO budget planning, and work packages for multi-million € EU projects
- Advise executives on AI strategy, sustainability, and emerging technologies; contributor to the ITU GDA Green Computing Working Group; frequent speaker and guest lecturer at industry events and universities
My mission: build AI that is not only powerful, but efficient, scalable, and sustainable—technically, economically, and environmentally.
I lead technology strategy and innovation across the tech directorate, advising executives on edge and cloud computing, autonomous networks, and AI/ML.
-
Autonomous Networks Journey (ANJ) — coordinate and report the program’s progress across networks, IT, security, sustainability, and CX to senior leadership locally and at group level, achieving an increase in autonomy from below 3 to 3.5+ (TM Forum methodology) in 2025, targeting Level 4 by 2030.
-
Large-scale budget planning — led total cost of ownership and business-case modeling for multi-million-euro infrastructure programs, across deployment and multi-year operations, with Controlling and commercial teams. Co-authored the €43M SPINE proposal to the EU’s IPCEI AI instrument, a consortium led by O2 Telefónica with Nokia, Capgemini, Fraunhofer, and TU Darmstadt.
-
European cloud & edge infrastructure — Project Coordinator for Telefónica Deutschland in EURO-3C, the Horizon Europe initiative for digital sovereignty spanning 85+ organizations across 13 countries. Also contributed to IPCEI-CIS, leading the deployment of the R&D edge node in Munich — coordinating cross-functional engineering teams and external vendors, and onboarding AI and AR/VR workloads.
-
Energy measurement and sustainability — led the redefinition of Telefónica Germany’s energy-consumption methodology, the basis for measuring and reporting the Tech ecosystem’s energy use, and its sustainability strategy with Tech and Corporate Responsibility, coordinating with Telefónica Global.
Energy-efficient AI consulting and tooling, helping SMEs use open-source models to cut the cost, energy, and emissions of AI inference. We benchmark combinations of models, APIs, and cloud or edge hardware, then deliver an actionable report showing where to reduce spend, keep data under the client’s control, and lower energy use.
Translating research findings on AI’s environmental footprint into actionable guidance for industry and regulators, drawing on work in LLM inference energy efficiency.
-
Led a research team to design and implement a system that analyzes application logs with Neural Temporal Point Processes and NLP, detecting errors and diagnosing their root causes. Patent pending in Germany and the US (DE102022131127A1, US20240176692A1).
-
Led the definition and implementation of GEC’s patenting process, with tech managers and the CTO.
-
Project Coordinator for the EU-funded “Model Learning for Cloud-Edge Digital Twins” project, managing budget and effort across the consortium, and led NEC’s contribution to the Horizon EU project BigDataStack (~€5M total consortium budget).
-
Developed a novel approach using reinforcement learning to automate the configuration and adaptation of data-driven applications. Recognized by the EU Innovation Radar, with three published papers and two granted patents in the US and EU.
Previous roles at NEC: Research Associate (October 2018 - July 2020) and Research Intern, Data Science and IoT (March 2018 - September 2018).
Developed features and improvements for Media5 Corp.’s VoIP iOS apps, including testing, issue fixing, and documentation.
-
Designed, implemented, and tested the Corrective Action Requests System, and led user training and issue fixing.
-
Implemented SyncroDB, a system to keep data consistent across different DBMS.
-
Designed, implemented, and tested the Personnel Attendance System.
I developed a novel algorithm capable of determining anchors in Tweets for automatic link generation under the supervision of Prof. Gill Dobbie. Funded by a Summer Research Scholarship from the University of Auckland.
Strategy & leadership: AI strategy · Innovation management · Strategic leadership · Technology roadmapping · EU research projects (Horizon Europe) · Executive advisory · IP and patenting processes
Sustainable computing: Energy measurement and profiling of software · LLM inference efficiency · Green AI metrics, benchmarks, and standards · AI environmental footprint reporting
AI/ML: Large Language Models · Reinforcement learning · Weakly supervised learning · Probability & statistics · Benchmarking and evaluation
Systems: Edge and cloud computing · Autonomous networks · Distributed systems · IoT · GPU inference serving
Tools: Python · PyTorch · HuggingFace · vLLM · SQL / NoSQL
- English [Fluent]
- Spanish [Native]
- German [Intermediate]
- Italian [Fluent]
Full list, including citation counts, on Google Scholar.
WattGPU: Predicting Inference Power and Latency on Unseen GPUs and LLMs
Watt Counts: Energy-Aware Benchmark for Sustainable LLM Inference on Heterogeneous GPU Architectures
Bench360: Benchmarking Local LLM Inference from 360°
VersaMatch: ontology matching with weak supervision
Tutor4RL: Guiding Reinforcement Learning with External Knowledge
Towards Knowledge Infusion for Robust and Transferable Machine Learning in IoT
Applying Weak Supervision to Mobile Sensor Data: Experiences with Transport Mode Detection
Reinforcement Learning Based Orchestration for Elastic Services
Elastic Services for Edge Computing
Towards adaptive actors for scalable iot applications at the edge
Data Applications and Operations
Granted and pending patents in the US, Europe, and Japan, filed at NEC Laboratories Europe and German Edge Cloud.
Automated knowledge infusion for robust and transferable machine learning
Weakly supervised reinforcement learning
Automated control through a traffic model
Ontology matching based on weak supervision
Method and system for supporting autonomous driving of an autonomous vehicle
Method and system for the analysis of application logs to detect errors and diagnose root causes
Data programming method for supporting artificial intelligence and corresponding system
Room-level indoor CO2 measurement without dedicated CO2 sensors
Open-source tooling from my research on the energy efficiency of AI, plus earlier personal projects.
Predicts the power draw and latency of LLM inference on GPUs and models it has never seen, using only publicly available model metadata and GPU specifications — no profiling or hardware access required. Median absolute percentage error of ≤3.4% for mean power draw on unseen GPUs in offline scenarios.
Watt Counts - Energy-aware LLM inference benchmark
The largest open-access dataset of LLM energy consumption: 5,000+ experiments across 50 LLMs and 10 NVIDIA GPUs in batch and server scenarios, with a reproducible open-source benchmark that accepts community submissions. Shows that GPU selection alone can cut inference energy by up to 70% in server scenarios with negligible impact on user experience.
EnergyMeter - Github repository
A Python library to measure the energy consumption of software, used to profile and optimize the efficiency of ML models and LLMs.
Bench360 - Local LLM inference benchmark
A framework to evaluate local LLM inference across tasks, usage patterns, and system metrics in one place — task quality alongside latency, throughput, energy, and startup time. Grew out of a Bachelor’s thesis I supervised.
Adaptive Applications Simulator - Github repository
A fast way to implement and test applications that adapt their logic to the current execution context, delivering high performance while meeting their requirements.
SIR on Gnutella - Github repository
Simulation of a Susceptible-Infected-Recovered Epidemic process on a Gnutella p2p network.
A photo editing application to create your own photo filters. No longer available on App Store.
Fixture 2014 - iOS Application
A World Cup Scorer to access live results from matches and the matches agenda. It reached #1 spot in Sports category of the App Store in Argentina and Uruguay. No longer available on App Store.
How much energy do LLMs consume?
We use EnergyMeter, a Python tool, to measure the energy consumption of different LLMs including Llama, Dolly, and BLOOM.
Transfer Learning in Reinforcement Learning
A quick review of how transfer learning improves the performance of RL on new, unseen tasks by exploiting learnings from past tasks.
5 Websites to Download Pre-trained Machine Learning Models
No need to train that machine learning model, just download a pre-trained one and let others do the heavy lifting!
How to Explain Decision Trees’ Predictions
We develop an approach to explain why a learned tree model chooses a certain class for a given sample, providing examples in Python.
Reinforcement Learning with TensorFlow Agents — Tutorial
Try TF-Agents for RL with this simple tutorial, published as a Google colab notebook so you can run it directly from your browser.
10+ Free Resources to Download Datasets for Machine Learning
A list of online resources to search and download datasets for your Machine Learning and AI projects
How to Access Stocks Market Data for Machine Learning on Python
If you want to use Machine Learning for trading stocks, you will need to create a dataset of stock markets data. Find out how to easily do it, with ready-to-use code.
Tutoring Reinforcement Learning
Reinforcement Learning agents start from scratch, knowing nothing and learning by experience, which is effective but slow. Could we give them some hints to get them started?
5 Frameworks for Reinforcement Learning on Python
Programming your own Reinforcement Learning implementation from scratch can be a lot of work, but you don’t need to do that. There are lots of great, easy and free frameworks to get you started in few minutes.
How is Reinforcement Learning used in Business?
Reinforcement Learning has proved it can achieve better results than humans in different games in recent years. But can RL also be used in businesses in the real world?
Entropy Regularization in Reinforcement Learning
In our everyday language, we commonly use the term “entropy” to refer to the lack of order or predictability of a system (for example, the universe.) In Reinforcement Learning (RL), the term is used in a similar fashion: in RL, entropy refers to the predictability of the actions of an agent.
Reinforcement Learning for everyone
RL has become popular in the AI community, but most people still don’t know what it is about. Come and read, no matter your background!
Make smarter agents with Hierarchical Reinforcement Learning
An introduction to Hierarchical Reinforcement Learning and an overview of different hierarchical approaches.
When to use Reinforcement Learning (and when not to)
What to consider to decide if Reinforcement Learning is the right approach to solve your problem.
Upcoming WattGPU: Predicting Inference Power and Latency on Unseen GPUs and LLMs (Paper Presentation)
AI & Digital in Action for NDCs: From Data to Impact (Panel)
An introduction to Large Language Models and their biases (Guest Lecture)
Cloud-Edge Continuum (Guest Lecture)
The Energy Efficiency of our Code [ESP: La Eficiencia Energética de Nuestro Código] (Seminar Presentation)
Challenges and Opportunities in Machine Learning [ESP: Desafíos y Oportunidades en Machine Learning] (Presentation)
Applying Weak Supervision to Mobile Sensor Data: Experiences with Transport Mode Detection (Paper Presentation)