Applied AI technical leadership · data science · production ML

Understanding complex systems through data and models.

I turn ambiguous problems into measurable data products, combining model-building depth with technical leadership, experimentation strategy, and cross-functional execution. I have guided production recommendation experiments at Mercado Libre, led an eight-person data science team at IThreex, and developed researchers through CONICET and twenty years of teaching at Universidad Nacional de Córdoba.

Download current CV Download one-page resume See selected projects

Millions of users Recommendation systems across Argentina, Brazil, and Mexico

3× training data Through pipeline, framework, and CUDA optimization

8-person DS team Technical roadmap, OKRs, mentoring, and client-facing delivery

30+ Q1 papers PhD, h-index 17, and international research collaborations

What I Bring

AI technical strategy

Problem framing, roadmap definition, model trade-offs, controlled experimentation, release risk, and business impact translated into practical technical direction.

Cross-functional leadership

Coordination across data science, engineering, product, analytics, clients, and academic teams, with clear communication of risks, next steps, and delivery constraints.

Evidence-driven validation

Statistical learning, Bayesian inference, A/B testing, confidence analysis, reproducible pipelines, and pre-production protocols for safer model releases.

Technical Leadership

Roadmaps and execution

Defined technical direction, contributed to OKRs, tracked delivery, and aligned AI work with client needs, KPIs, and operational constraints.

Experiment strategy

Guided production experiment decisions by balancing statistical evidence, business impact, cost, model behavior, and release risk.

People and evaluation

Mentored and onboarded data scientists, supervised doctoral researchers, reviewed technical work, and participated in hiring and evaluation decisions.

Interactive Data Science Lab

Technical notes you can manipulate

Small interactive applets showing how metrics, projections, uncertainty, and model behavior change as assumptions change.

ROC and PR Explorer

Classification thresholds, class imbalance, and the gap between ROC AUC and Precision-Recall behavior.

PCA Projection Explorer

A geometric view of variance, projection, reconstruction, and what lower-dimensional representations preserve.

Selected Experience

2024-2026

Senior Data Scientist / Machine Learning Engineer, Mercado Libre

Primary data science owner for an online ranking model used by dozens of internal clients and millions of users across three LATAM markets.

Scope and outcomes

Extended a PyTorch ranking model with new features, embeddings, and production feature-store data.
Tripled training-data capacity by optimizing Fury pipelines, framework configuration, and CUDA.
Introduced MLflow tracking and a pre-production validation protocol; prepared ONNX artifacts and partnered with engineering on reliable releases.
Built confidence tools for A/B tests and monitored clicks, purchases, GMV, and model behavior in Datadog and Looker.
Guided launch, pause, and refinement decisions by communicating evidence, trade-offs, risks, and next steps to product, engineering, analytics, and leadership stakeholders.
Delivered measurable purchase-conversion lift through controlled production experiments.

2022-2024

Lead Data Scientist, IThreex Global

Led eight data scientists delivering client-facing AI and analytics across public revenue, retail, tourism, agriculture, and international trade.

Scope and outcomes

Defined the technical roadmap for the Molibdeno AI platform, prioritizing reusable ML workflows, client needs, and delivery constraints.
Defined OKRs, tracked execution, mentored and onboarded data scientists, and participated in hiring and evaluation decisions.
Acted as principal technical contact for Kolektor, delivering payment-behavior analysis and tax-revenue forecasting.
Delivered segmentation and predictive work supporting commercial strategies for Pueblo Nativo, Inverfin, and Agencia ProCórdoba.
Built computer-vision models and image-acquisition protocols for cattle-weight estimation, plus a LangChain RAG service exposed through an API.
Translated requirements into data products, presented results to clients, and mentored team members in active projects.

2023-present

Associate Professor, Universidad Nacional de Córdoba

Head of the Data Science course in Applied Mathematics, building on twenty years teaching statistics, machine learning, and scientific computing.

Academic leadership

Designs and teaches technically rigorous material, supervises graduate researchers, and connects mathematical foundations with practical data-science implementation.

2004-2023

Researcher, CONICET

Applied Bayesian inference, statistical learning, numerical methods, and HPC to large astronomical datasets in international, multidisciplinary teams.

Research foundation

Developed open-source scientific software and automated analysis pipelines; authored 30+ articles in international Q1 journals, supervised researchers, reviewed technical work, and organized academic activities.

Marcelo Lares