Hello, I'm

Hugo Gobato
Souto

From self-taught researcher to leading causal AI at scale — driven by curiosity, shaped by two cultures.

0 Citations
0 Publications
0 Languages
Hugo Gobato Souto
Ilhabela, Brazil

// About Me

I am a research scientist working at the intersection of Causal Inference, Agentic AI, and Quantitative Finance. Currently serving as Principal Causal AI Research Scientist at Dell Technologies, I coordinate research on the synergies of Agentic AI, Causal Inference, and Causal Reinforcement Learning.

My academic work has been published in journals such as Expert Systems with Applications, Financial Innovation, Computational Economics, and The Journal of Finance and Data Science. I have also contributed to the GPT-4o System Card as part of the OpenAI Red Teaming Network.

Born in Brazil, shaped by the Netherlands — I spent formative years studying and working in the Dutch academic and professional world, an experience that deeply influenced how I think and work. I speak five languages and am always open to new international collaborations and opportunities.

Research Interests

Causal Inference Agentic AI Causal Reinforcement Learning Quantitative Finance Realized Volatility Deep Learning Topological Data Analysis

Languages

PortugueseNative (C2)
EnglishC2
FrenchC2
DutchC2
SpanishB2

Open to Relocation

Currently based in Ilhabela, Brazil. Eager to work abroad again and open to relocating internationally.

🤖

Agentic Synergy: Why Causal RL?

Traditional Reinforcement Learning often fails in complex environments because it confuses correlation with causation. By integrating Causal Inference, my research enables agents to perform counterfactual reasoning. Instead of just maximizing rewards based on what they see, they ask: "What would have happened if I had taken a different action?" This allows for breaking confounding bias ($U$) and achieving true sample efficiency.

Causal RL: The Interventional Agent

Unlike passive AI, a Causal Reinforcement Learning agent doesn't just observe correlations—it performs $do(X)$ to break confounding bias and learn the true structure of the world.

Agent Mode: Observational

\( E[Y|X] \)


> Initializing environment...
> Conflicting signals from U detected.

// Explore in Depth

The headlines are above. If you want the full record — every paper, every letter, every talk — it lives on these pages.

// My Story

A path built on curiosity, grit, and challenging the status quo.

2019

The Leap

At 17, after a year of high school in the Netherlands because of my mom's Post Doc, I fell in love with the country and its culture. When my family's visa expired, I returned to Brazil — but I knew I had to find a way back. I applied for scholarships and was awarded both the Holland Scholarship and the HAN Excellence Scholarship to study International Business at HAN University of Applied Sciences in Arnhem.

Rivers International School Arnhem 2018/2019 class photo Rivers International School Arnhem, 2018/2019
Sunset over the Rhine in Arnhem Arnhem — the city that became home
2020–22

Teaching to Survive, Learning to Thrive

With only €500/month from my parents and living costs of €1,200, I needed a job. After 35+ rejections (most due to work permit complications), I landed a position as a Portuguese teacher at Volksuniversiteit Zwolle — which grew into teaching French, English, and Dutch as well, plus private lessons. I taught myself Dutch and French independently to professional fluency.

During academic terms, I worked approximately 40 hours per week on top of my full-time studies. During vacations, this climbed to 80 hours, with up to 65 one-hour lessons per week. Through it all, I maintained an average grade of 8.85 out of 10.

Portuguese course listing at Volksuniversiteit Zwolle taught by Hugo Gobato Souto Portuguese course at Volksuniversiteit Zwolle
2022–23

The Research Breakthrough

Even though I was at an applied university (HBO), I felt a deep drive to do research. With no formal training in advanced mathematics, I taught myself through online courses from Columbia, UC, and HKUST. Then I started publishing — entirely on my own initiative.

During my Bachelor's alone, I published 7 research papers, including one proposing a novel financial theory using Topological Data Analysis — the Topological Tail Dependence Theory, which connects persistent homology to stock market behavior. This caught the attention of Professor Amir Moradi, who became my research mentor and collaborator.

Meanwhile, at BASF Nederland, I managed financial risk for the Hollandse Kust Zuid project — the world's largest offshore wind farm — using advanced statistical models.

Presenting at the 16th International Risk Management Conference in Florence Presenting at the Risk Management Conference, Florence
With the BASF Nederland team With the BASF Nederland team
2023

Recognition

The hard work paid off. I was named HAN Student of the Year 2023 and won the Best Bachelor Thesis Award. I was also invited to join the OpenAI Red Teaming Network, contributing to the safety evaluation of models like GPT-4o. I presented my research at conferences in Florence, Glasgow, London, and Barcelona.

Receiving the HAN Prijs Student of the Year 2023 HAN Student of the Year 2023
Poster presentation at ESOBE 2023, Glasgow Poster at ESOBE 2023, Glasgow
2024–25

New Chapter: Causal Inference

I decided to return to Brazil to be closer to family, and pivoted my research to Causal Inference. I enrolled in a double-degree Master's in Statistics at USP and UFSCar, earning both the FAPESP and CAPES scholarships.

I quickly became an expert in the field, writing three research papers during my Master's and finishing the program ahead of schedule (September 2025 instead of March 2026). In industry, I led causal experimentation at Luizalabs (Magazine Luiza), one of Brazil's largest e-commerce companies.

A toucan at USP campus, Sao Paulo Wildlife at USP — a toucan greeting at campus
Presenting causal inference research at EUROCIM 2025 Presenting causal inference research, EUROCIM 2025
2026

Today & Beyond

Now as Principal Causal AI Research Scientist at Dell Technologies, I'm exploring the frontier where Agentic AI meets Causal Inference and Causal Reinforcement Learning. With 295 citations at age 24, I continue to push boundaries — always looking for the next problem to solve, the next assumption to challenge.

Hugo's research workspace Where the research happens
Ilhabela, Brazil — Hugo's current home Ilhabela, Brazil — home base

I'm open to international opportunities and always eager to collaborate on research that challenges the status quo.

// Professional Experience

Principal Causal AI Research Scientist

Dell Technologies Feb 2026 — Present

Coordinating research in the synergies of Agentic AI, Causal Inference, and Causal Reinforcement Learning.

Team-Leading Causal Data Scientist

Luizalabs (Magazine Luiza) Feb 2025 — Feb 2026

Led coordination of e-commerce experiments and extracted business value through causal inference methodologies.

Researcher — Red Teaming Network

OpenAI Nov 2023 — Jul 2025

Contributed to improving the safety and effectiveness of OpenAI's models, including work on the GPT-4o System Card.

Research & Teaching Assistant

HAN University of Applied Sciences Jul 2023 — Jan 2024

Research on novel neural networks and sustainability in businesses and finance within the International Business faculty.

Project Financial Analyst

BASF Nederland Aug 2022 — Jul 2023

Managed and mitigated project risks of Hollandse Kust Zuid (world's largest offshore wind park) using state-of-the-art statistical and stochastic models.

Language Teacher

Volksuniversiteit Zwolle + Private Lessons Aug 2020 — Aug 2022

Taught French, English, Dutch, and Portuguese — 30 to 65 lessons per week while pursuing a full-time Bachelor's degree.

// Education

My master's dissertation and its three methodological contributions are described on the Research page.

ESG Portfolio Management & Mathematical Finance

ICHEC Brussels School of Management

Feb 2022 — Jul 2022

Financial Engineering & Risk Management

Columbia University (Online)

Nov 2022 — Jan 2023

Advanced Bayesian Statistics

University of California (Online)

Nov 2022 — Jan 2023

Mathematics for Engineers

HKUST (Online)

Oct 2022 — Nov 2022

// Selected Publications

Google Scholar · ORCID — 295 citations · h-index: 9 · i10-index: 9 · see the complete record →

Topological Insight: Persistent Homology

My research uses Topology to find "holes" in financial data that traditional statistics miss. As the connection radius increases, simplicial complexes form, revealing the underlying shape of market risk.

2026

A Mathematical Framework for Topological Causal Data Analysis

H.G. Souto, I. Diamantis

arXiv:2607.28161 [stat.ME] — read the abstract

Work in Progress
2024

GPT-4o System Card

OpenAI — technical report

Contributed safety evaluations as a member of the OpenAI Red Teaming Network (not listed as an author)

2025

AI-PAVE-Br: Leveraging LLMs for Enhanced Product Attribute Value Extraction

M. Gazzola, H. Souto, S. Silva, J. Peixoto, F. Siqueira, A. Morais, C. Gomes

Anais do XVI STIL, pp. 149-160

2025

Enhancing Financial Forecasting: A Novel Multivariate Neural Network Approach Using RCM-NBEATSx

H.G. Souto, A. Moradi

Financial Innovation, 11 (Accepted, Dec 2025)

2025

Evaluating the Efficacy of NHITS for Forecasting Stock Realized Volatility

H.G. Souto

Computational Economics, 65(3)

7 citations
2024

Charting New Avenues in Financial Forecasting with TimesNet

H.G. Souto

Expert Systems with Applications, 255

16 citations
2024

Can Transformers Transform Financial Forecasting?

H.G. Souto, A. Moradi

China Finance Review International (Special Issue: AI & Finance)

33 citations
2024

Time-mixing and Feature-mixing Modelling for Realized Volatility Forecast

H.G. Souto, S.K. Heuvel, F. Louzada Neto

The Journal of Finance and Data Science, 10

11 citations
2023

Introducing NBEATSx to Realized Volatility Forecasting

H.G. Souto, A. Moradi

Expert Systems with Applications, 242

54 citations
2023

Topological Tail Dependence: Evidence from Forecasting Realized Volatility

H.G. Souto

The Journal of Finance and Data Science, 9

23 citations
2023

A Novel Loss Function Using Wasserstein Distance

H.G. Souto, A. Moradi

Decision Analytics Journal, 10

18 citations
2023

Forecasting Realized Volatility through Financial Turbulence and Neural Networks

H.G. Souto, A. Moradi

Economics and Business Review, 9(2), 134-159

21 citations
2025

Forests for Differences: Robust Causal Inference Beyond Parametric DiD

H.G. Souto, F. Louzada Neto

arXiv preprint

2 citations Work in Progress
2024

Beyond Arbitrary Replications: A Principled Approach to Simulation Design in Causal Inference

H.G. Souto, F. Louzada Neto

arXiv preprint — introduces TISCA

Work in Progress
2024

Ablation Studies for Novel Treatment Effect Estimation Models

H.G. Souto, F. Louzada Neto

arXiv preprint

3 citations Work in Progress
2024

Advancing Causal Inference: Nonparametric ATE and CATE Estimation with Continuous Treatments

H.G. Souto, F. Louzada Neto

arXiv preprint

5 citations Work in Progress

// In Their Words

A few excerpts from the reference letters. Full letters are shared on request, with the author's permission.

Master's Supervisor
Mr. Hugo Gobato Souto is in the top 2% of the most capable researchers I have supervised out of the dozens of students I have taught or supervised at USP.
Prof. Francisco Louzada Neto, D.Phil. Full Professor of Statistics, University of São Paulo; Director, CeMEAI
Independent Peer
It was not the defence of an author who needed his paper to be right, but the engagement of a researcher who appeared at least as curious as I was about whether his own result would survive a serious challenge.
Yusufhan Balci Doctoral Researcher in Epidemiology and Data Science, Amsterdam UMC
Research Collaborator
He is unusual because he can connect these areas in a way that is mathematically informed, statistically careful, and computationally realistic.
Ioannis Diamantis, Ph.D. Assistant Professor, School of Business and Economics, Maastricht University
Industry Colleague
It is uncommon to find someone who is genuinely strong across analysis, agentic AI systems, and software engineering at the same time. […] Hugo moves fluently between all three.
Pedro Henrique Braga Lisboa Principal Research Scientist, Dell Technologies

Read all 36 endorsements

// Awards & Honors

Dates, awarding bodies, and scholarship reference numbers are listed on the Academic Service page.

2023

HAN Prijs — Student of the Year

HAN University of Applied Sciences

Hugo receiving the HAN Prijs Student of the Year 2023 award
2023

Best Bachelor Thesis of the Year

HAN University of Applied Sciences

Certificate for Best Graduation Research Report Award
2024

FAPESP Master's Scholarship

FAPESP, Brazil

2024

CAPES Master's Scholarship

CAPES, Brazil

2023

Distinguished Nomination — ECHO Award

ECHO

2019

HAN Excellence & Holland Scholarship

HAN University / Holland Scholarship Programs

// Conference Presentations

A selection. The full list, together with invited university lectures, podcasts, and press coverage, is on the Talks & Media page.

WPSM 2026

"On the Applications of Nonparametric Bayesian Methods in Heterogeneous Treatment Effect Estimation"

Federal University of São Carlos, Brazil · February 2026

CDSM 2025

"Forests for Differences: Robust Causal Inference Beyond Parametric DiD"

Causal Data Science Meeting · Online · November 2025

Presenting at EUROCIM 2025, University of Ghent

EUROCIM 2025

"Revisiting Model Evaluation in Treatment Effect Estimation"

University of Ghent, Belgium · April 2025

CIICE

"To What Extent Does ESG Impact Profitability Within the Airline Industry"

University of Barcelona, Spain · November 2023

Poster at ESOBE 2023, Glasgow

ESOBE 2023

"Application of Persistent Homology in Forecasting Realized Volatility"

University of Glasgow, Scotland · August 2023

At the 16th International Risk Management Conference, Florence

16th International Risk Management Conference

"NBEATSx Is All What You Need for Realized Volatility Forecasting"

University of Florence, Italy · July 2023

Presenting at the 25th Dynamic Econometrics Conference

25th Dynamic Econometrics Conference

"Realized Volatility Predictability through Neural Network and Financial Turbulence"

Bayes Business School, UK · April 2023

// Technical Skills

Programming & ML

Python
PyTorch
TensorFlow
R
C++

Data & Analytics

SQL
MATLAB
SAP

Tools & Writing

LaTeX
Office 365

Research Domains

Causal Inference Causal Reinforcement Learning Realized Volatility Forecasting Topological Data Analysis Time Series Forecasting Treatment Effect Estimation Financial Turbulence

// Community

Teaching, thesis committees, and peer review are on the Academic Service page.

// Get in Touch

Open to research collaborations, international opportunities, and interesting conversations.

Let's talk

For research collaborations or business consultations — book an initial call so I can understand your business problem, see if I can be useful, and we can agree on a consultation price together for the project.

Book a time for a call

Free initial call · price agreed together if we proceed · via Google Calendar

Research Assistant

Hugo's Agentic AI Simulation

Hello! I'm an Agentic AI simulation trained on Hugo's research and experience. How can I help you today?