1. 01About
  2. 02Experience & Education
  3. 03Tech Stack
  4. 04Projects
  5. 05Profile

Mario Gertrudix Moreno

About

Based in Madrid. I have a background in applied statistics
and a mix of corporate and entrepreneurial experience.
I like data-driven decisions, new technology
and rethinking the things everyone takes for granted.

Experience

  • Business Operations Analyst (Internship)

    Repsol · Jun 2024 – Jul 2025

    Performance department for Lubricants, Aviation, Asphalts and Specialties: process automation, the 2025 budget and cost analysis.

    Process automation

    Month-end close reporting for four business verticals automated with VBA macros (SAP and Excel), run overnight in Automation Anywhere: around 12 person-days of manual extraction a month replaced.

    No more month-end overtime for a three-person team and no manual transcription errors in a process unchanged for almost ten years, built together with the team that ran it.

    Ownership and escalation layer of a daily sanctions-screening alert system: every product family and TARIC code mapped to its sales owner across ~30 commercial contacts, risk-based prioritisation, and an AI-driven flow that reads image-based carrier invoices and prepares a ready-to-send case file for Legal.

    Forecasting and budgeting

    2025 budget model built from scratch, with time series models per market chosen by backtesting against prior years; adopted by the team.

    ~3% error in large markets and ~4% aggregate error across verticals; line-level budget deviation cut from ~17% to ~12%, with nearly 30% fewer manual reallocations during the year.

    Cost analysis

    Cost analysis in SAP to identify non-essential spend and keep every vertical within budget.

  • Freelance

    Dataevidence.tech · Dec 2021 – Present

    Websites and corporate identities for institutional clients, from requirements to delivery, and custom suites built around each client's needs.

Education

  • Bachelor's Degree in Applied Statistics

    Complutense University of Madrid · Sep 2022 – May 2026

    Undergrad thesis on prediction markets.

  • HKU Summer Institute

    The University of Hong Kong · Jul 2023

    Big Data & Analytics in Business.
    Sustainability Leadership for a Circular Economy.

  • Summer Academic Experience

    University of St Andrews · Jul 2019

    Physical Science, Mathematics, Computer Science.

Certifications

  • AWS Certified Cloud Practitioner

    Amazon Web Services (AWS) · Nov 2025

    +
  • Foundry & AIP Builder Foundations

    Palantir Technologies · Aug 2025

    +
  • SAS – Complutense University of Madrid Academic Specialization in Programming

    SAS · Jul 2023

    +

Tech Stack

Projects

Oct 2026

Demand forecasting, supply planning and sales support for Spanish jet fuel and lubricants, built on CORES, CNMC and Aena data and deployed on AWS.

54series forecast every month

  • AWS
  • SageMaker
  • Bedrock
  • LightGBM
View projectLive site ↗

Aug 2026

Open comparator of academic outcomes (pass, dropout and graduation rates) of Spanish university degrees, down to subject level.

7.8M+observations from 41 sources

  • Python
  • DuckDB
  • dbt
  • OCR
View projectLive site ↗

Bachelor's thesis · Jun 2026

How well prediction markets anticipate the Fed's rate decisions, against the institutional tools, across thirteen FOMC meetings.

3.18bplowest average error: Polymarket

  • Python
  • Diebold–Mariano
  • Bootstrap
  • Granger
View projectThesis PDF ↗

Jet Fuel & Lubricants Demand Planning Platform on AWS · Oct 2026

The data to plan Spanish jet fuel and lubricants lives in separate CORES, CNMC and Aena releases. The platform brings them together on AWS and turns them into demand forecasts, a supply programme, margins and answers for the sales team.

  • A serverless pipeline (EventBridge, Step Functions, Lambda) loads every new CORES and Aena release into Iceberg tables on S3, with Glue and 21 Athena views, keeping each data vintage for leakage-free backtests.
  • A monthly SageMaker Processing job picks a champion model per series by rolling backtest: SARIMA/SARIMAX, global LightGBM, Timestral or TimeGPT.
  • A sales agent on Bedrock AgentCore (Claude via Bedrock) answers salespeople from the same Athena data, traced in CloudWatch and X-Ray.

54series

forecast every month, together with a linear supply programme and margins.

4

model families compete for each series in a rolling backtest; the best becomes its champion.

How it works

  1. 01Sources

    CORES · CNMC · Aenaseparate data releases
  2. 02Ingestion

    Serverless pipelineEventBridge, Step Functions and Lambda
  3. 03Storage

    Iceberg tables on S3Glue and 21 Athena views, every data vintage kept
  4. 04Outputs

    Monthly forecast runSageMaker picks a champion per series: 54 forecasts, supply programme and margins
    Sales agentBedrock AgentCore, answering salespeople from the same data

comparatucarrera.com · Aug 2026

Pass, dropout and graduation rates of Spanish degrees are scattered across quality reports, open-data portals and PDFs. The comparator brings them together down to subject level, with every figure traceable to the page it came from.

  • A pipeline in Python, DuckDB/Parquet and dbt (bronze, silver, gold) over 41 sources, with OCR for PDFs. Each observation keeps its printed literal, page and document hash.
  • Arithmetic-identity checks and 57 dbt tests, with degree-plan lineages and double degrees modelled on the RUCT registry to avoid double counting.
  • Developed with AI coding agents under task specs, a repo contract (30+ ADRs) and an automated validation harness, with manual review on every published figure.

7.8M+

observations from 41 sources.

≈0pp

median deviation on core rates against the official benchmarks (SIIU, QEDU, RUCT).

How it works

  1. 01Sources

    41 sourcesquality reports, open-data portals and PDFs
  2. 02Extraction

    Python and OCReach value keeps its printed literal, page and hash
  3. 03Modelling

    DuckDB and dbtbronze, silver and gold layers, with RUCT lineages
  4. 04Validation

    57 dbt testsidentity checks and official benchmarks: SIIU, QEDU, RUCT
  5. 05Output

    comparatucarrera.comrates by degree, down to subject level

Bachelor's thesis · Complutense University of Madrid · Jun 2026 · Supervised by Enrique González Arangüena

Prediction markets such as Polymarket and Kalshi now trade at scale on the Fed's rate decisions. The thesis tests how they compare with CME FedWatch and Fed Funds Futures on accuracy, calibration and speed of response, over the thirteen FOMC meetings from June 2024 to December 2025.

  • Trade-level data turned into daily VWAP series, read as four outcomes and as an expected rate (bp).
  • Brier score, RPS, MAE and RMSE, with Diebold–Mariano tests on cluster-robust errors and a 5,000-replicate cluster bootstrap.
  • Panel Granger causality to see which source moves first.

3.18bp

Polymarket's mean absolute error, the lowest of the four (futures read directly: 8.04 bp).

11 of 12

meetings where FedWatch's binomial tree significantly beats the raw futures. The futures lead price discovery; Polymarket leads Kalshi.

How it works

  1. 01Sources

    Polymarketon-chain trades
    Kalshitrades
    CME FedWatchdaily probabilities
    Fed Funds Futuresdaily settlements
  2. 02Series

    Daily seriesVWAP for the markets; 13 FOMC meetings, read as four outcomes and as an expected rate
  3. 03Analysis

    Accuracy and calibrationBrier, RPS, MAE, RMSE; Diebold–Mariano and cluster bootstrap
    Price discoverypanel Granger causality
  4. 04Findings

    Lowest error: Polymarket3.18 bp MAE; the futures lead price discovery

Profile

Where I fit best.

Bringing data, decisions and operations together.

Illustrative k-nearest-neighbours map with simulated profiles. Horizontal axis: business to technical work; vertical axis: analysing and advising to building and running. Of the seven profiles closest to this one, two work in business operations, two are deployment strategists, two work in data analysis and one is a forward-deployed engineer.

Contact

[email protected] LinkedIn ↗