Decision Scientist | Semantic Layer Governance, Gen AI Evals, and Forecasting

I build the layer a business self-serves through, and I defend what the numbers in it mean. On contract with the central data team at Kit, that means co-owning our codebase — semantic layer and dbt marts, shipped as reviewed PRs — and holding one definition per metric against every request to fork it.

The building is the easy half. The hard half is arbitrating between two teams who each want the metric measured their own way, and evals on Gen AI output, because a stakeholder cannot tell a right answer from a confident wrong one and the model will hand them either.

Before that, nineteen years across finance, operations and data with a forecast to defend in almost every one of them: rolling forecasts and budgets at Philip Morris International and Colgate-Palmolive, now MRR and ARR at Kit.

Since September 2026 I also teach the subject, as Adjunct Professor of Time Series Analysis and Forecasting at IE University.

Working on

• Semantic layer governance — topic scoping, metric definitions, and the AI context an LLM needs before it can be trusted to query a model
• Single source of truth as a practice, not a slogan: one definition, one owner, one place to change it
• Evals on Gen AI analytical output: right topic, right joins, right filters, right grain
• Revenue and ARR/MRR forecasting, and the targets that hang off it
• A/B test design, and the predictive work that has no other home on the team

Adjunct Professor in the Dual Degree in Business Administration & Data and Business Analytics. Course materials are public and taught in R on the tidyverts stack, following Forecasting: Principles and Practice, 3rd ed.

The method is build it by hand, then verify against the library. Students implement classical decomposition, simple exponential smoothing, prediction intervals and the error metrics themselves before calling the function that does them.

• SQL: Redshift, DuckDB, BigQuery
• Python: pandas, scikit-learn, statsmodels, plotnine, Plotly (uv for dependencies)
• R: tidyverts — tsibble, feasts, fable — and the tidyverse
• Transformation: dbt
• Modeling: time series forecasting (ETS, ARIMA), MRR forecasting, attribution heuristics
• Semantic layer: Omni topics, dbt marts, metric definitions, AI context fields
• BI: Omni, Count, Mode, Preset (Apache Superset), Metabase, Tableau, QuickSight
• AI: AWS Bedrock, eval harnesses for analytical LLM output
• Tooling: VS Codium, DBeaver, Podman, Fedora

• AWS Certified AI Practitioner (AIF-C01)
  • Validate at https://aws.amazon.com/verification; verification code d6caf1808e55484eb84ac24ce254d101
• BetterWorks OKR Master
• Data Storytelling & Visualization (The Economist)

• Email aleberriz@pm.me
• LinkedIn https://linkedin.com/in/berriz
• Website https://aleberriz.com
• X @alejandroberriz
• ORCID https://orcid.org/0009-0008-9811-9822

Trilingual: English, Spanish, French.