Contato

Eirik R.
[Link]/in/eric-arayai AI & Data Science | Generative AI & Agents | Geometric &
(LinkedIn) Representation Learning | LLM Research | Hyperbolic Intelligence
[Link]/ArayAI/ (Project Author) | Independent Researcher
[Link] (Company) São Paulo, Brasil

Principais competências
Resumo
Predictive Analytics & Demand
Forecasting I am an Artificial Intelligence and Data Science specialist with
Statistical Modeling & A/B Testing over nine years of experience developing, deploying, and scaling
Customer Segmentation & Marketing Machine Learning and Generative AI models. My expertise includes
Analytics
MLOps, Predictive Modeling, Computer Vision, and Large Language
Models (LLMs), applied across the finance, renewable energy, and
Languages
biometrics industries.
Inglês (Professional Working)
Português (Native or Bilingual)
I have strong experience in cloud-based AI architectures, real-
time anomaly detection, and predictive system optimization
Certifications
using platforms such as AWS, Azure, and Databricks, alongside
Multi AI Agent Systems with crewAI
orchestration tools including Docker, Kubernetes, Kedro, and Airflow.
Retrieval Augmented Generation for
Production with LangChain &
My work consistently combines advanced research with practical
LlamaIndex implementation, delivering scalable, data-driven solutions with
Domine LLMs com Langchain measurable business impact.
Practical Multi AI Agents and
Advanced Use Cases with CrewAI
I am the author of Hyperbolic Intelligence, an open research
initiative exploring non-Euclidean geometries applied to AI. This
Publications
project investigates hyperbolic embeddings, hierarchical semantic
Dimensional Efficiency Bounds
representations, and the scaling limits of large language models,
for Embedding Hierarchical Metric
Structures: A Regime-Dependent focusing on how geometric structures beyond traditional vector
Analysis spaces can improve semantic organization, generalization, and
Hyperbolic Neural Cellular Automata: interpretability in intelligent systems. The work bridges theoretical
A Geometric Framework for
Emergent Complexity research and experimental prototyping and is developed as an open,
The Topological Signature of research-first initiative.
Consciousness: A Gromov-
Wasserstein Framework for Neural-
Phenomenal Alignment In parallel, I am pursuing a Bachelor’s Degree in Psychoanalysis,
Before the Model: Decision and deepening my understanding of human behavior, unconscious
Responsibility in Data Projects That dynamics, and the intersections between mind, culture, and
Must Survive the Real World
technology. This interdisciplinary perspective informs my AI work by
Hyperbolic Semantic
Communication: The helping me design systems that better account for cognitive biases,
Geometrodynamics of Synthetic emotional responses, and human-centered constraints—enabling
Intelligence under Causal and
more ethical, robust, and effective solutions.
Perceptual Constraints
Page 1 of 8
I bring a strong technical foundation, hands-on leadership
experience, and a results-oriented mindset. I value teams and
organizations that prioritize technical excellence, intellectual honesty,
innovation, and long-term impact.

I am committed to working exclusively with organizations that

uphold ethical and transparent cultures and actively value human
development. My focus is on opportunities where advanced AI
can generate sustainable, positive impact while supporting the
responsible adoption of intelligent systems.

My purpose is to combine technological innovation with ethical

responsibility, contributing to a more fair, transparent, and human-
centered future.

Experiência
DHARMA-AI
Senior Data Scientist
março de 2026 - Present (1 mês)
I work at Dharma AI in the Research and Development (R&D) area,
contributing to the creation of innovative solutions based on Small Language
Models (SLMs).

My work involves validating technical hypotheses, developing AI pipelines, and

building specialized datasets for model training and evaluation. I operate within
agile experimentation cycles, transforming real-world problems into scalable
solutions with a strong focus on performance, cost-efficiency, and intellectual
property generation.

I also contribute to improving the company’s “model factory” — an ecosystem

of internal tools that automates key stages such as data generation, training,
inference, and evaluation. Additionally, I work on model performance analysis,
technical documentation of experiments, and knowledge dissemination across
the organization.

Page 2 of 8
My role bridges applied research and product development, ensuring that
technical advancements are translated into practical and efficient business
solutions.

Arayai
Research Scientist
julho de 2024 - Present (1 ano 9 meses)
Leading exploratory research on non-Euclidean foundations of intelligence,
with a focus on hyperbolic representation learning, geometric compression,
and dynamical systems for scalable semantic modeling.

Author and architect of the Contrastive Geometric Transfer (CGT) framework,

investigating whether negatively curved spaces can compress high-
dimensional Euclidean embeddings while preserving semantic structure.

Research Focus

• Hyperbolic embedding compression

• Contrastive geometric learning
• H-AKORN (Hyperbolic Attentive Kuramoto Oscillator Networks)
• Neural Cellular Automata in non-Euclidean manifolds
• Topological Data Analysis (persistent homology, Betti numbers)
• Geometric loss functions and curvature-aware optimization

Bridging differential geometry, dynamical systems, and deep learning to

explore alternative substrates for intelligence beyond brute-force scaling.

Alternative Version (More Strategic Positioning)

Senior Research Data Scientist | Geometric Intelligence & Representation
Learning

Researching next-generation AI architectures grounded in differential

geometry and dynamical systems.

My work challenges the assumption that intelligence scales solely with

parameters and compute, exploring whether the geometric structure of
representation spaces can unlock more efficient and expressive models.

Key Contributions

Page 3 of 8
• Developed CGT (Contrastive Geometric Transfer) for hyperbolic embedding
compression
• Implemented falsifiable geometric intelligence simulations (UGFT framework)
• Designed H-AKORN phase dynamics for synchronization-driven
representation learning
• Integrated topological constraints into neural optimization pipelines

FIT - Instituto de Tecnologia

Senior Data Scientist
dezembro de 2022 - julho de 2024 (1 ano 8 meses)
At FIT, I develop and deploy AI-driven solutions to optimize solar tracking
systems and battery lifecycle forecasting, enhancing operational efficiency and
predictive maintenance. Highlighted Projects & Contributions:

Developed predictive models for solar tracking optimization, reducing

unexpected failures and improving system efficiency.

Built battery lifecycle forecasting models, leveraging time series analysis and
degradation modeling.

Implemented real-time anomaly detection systems for predictive maintenance

and performance monitoring.

Established MLOps workflows, deploying scalable AI solutions in AWS using

Docker, Kubernetes, and Lambda functions.

Automated large-scale data pipelines with Airflow and Kedro, streamlining data
processing in cloud environments.

Paraná Banco S/A

Senior Data Scientist
agosto de 2021 - julho de 2022 (1 ano)
At Paraná Banco, I worked on AI-powered risk assessment models, NLP-
based customer analytics, and Open Finance strategies. Highlighted Projects
& Contributions:

Developed credit risk assessment models, optimizing credit scoring, fraud

detection, and loan approval automation.

Page 4 of 8
Designed an NLP-driven system to classify SMS customer responses,
enabling automated segmentation and analysis in Power BI.

Created churn prediction and lifetime value (LTV) forecasting models,

enhancing customer retention strategies.

Automated AI workflows in Databricks, utilizing PySpark and cloud-based AI

solutions to handle high-volume data processing.

Natosafe
Senior Data Scientist
dezembro de 2019 - julho de 2021 (1 ano 8 meses)
At Natosafe, I focused on biometric AI and fingerprint recognition technologies,
developing secure identification solutions for infant biometrics. Highlighted
Projects & Contributions:

Developed AI models for infant fingerprint recognition, addressing challenges

in biometric identification for children aged 0-5.

Implemented deep learning-based segmentation networks, improving the

accuracy of fingerprint analysis.

Applied super-resolution AI techniques to enhance biometric image quality for

secure authentication.

Deployed cloud-based biometric AI models, ensuring scalable and efficient

processing of biometric data.

MadeiraMadeira
Senior Statistical Analyst
dezembro de 2018 - novembro de 2019 (1 ano)
Curitiba, Paraná, Brasil

At MadeiraMadeira, I focused on predictive analytics, customer segmentation,

and inventory forecasting. Highlighted Projects & Contributions:

Developed demand forecasting models, optimizing inventory management and

reducing operational costs.

Built AI-driven clustering models for customer segmentation, improving

personalized marketing strategies.

Page 5 of 8
Designed and implemented a ranking algorithm for product recommendations,
enhancing conversion rates.

Automated business intelligence reporting using Tableau, AWS, and Airflow.

FORTBRASIL Administradora de Cartões

Credit Risk Analyst
fevereiro de 2016 - abril de 2017 (1 ano 3 meses)
Fortaleza e Região, Brasil

At FortBrasil, I played a key role in monitoring credit risk models, conducting

A/B testing, and tracking model performance. Highlighted Projects &
Contributions:

Conducted A/B testing on credit risk strategies, optimizing approval and fraud
detection rates.
Monitored and evaluated third-party machine learning models, providing data
for continuous retraining and performance improvement.
Analyzed model drift and rolling performance, ensuring long-term stability and
compliance with risk regulations.

FUNCAP
Scholarship Holder
fevereiro de 2015 - fevereiro de 2016 (1 ano 1 mês)
Fortaleza e Região, Brasil

As a research scholar funded by FUNCAP, I developed an AI-powered

intelligent people counting system using computer vision and meta-heuristics,
aimed at real-time crowd monitoring and movement analytics. This project
played a critical role in urban planning, security enhancement, and business
intelligence applications, leveraging advanced image processing, machine
learning, and state estimation models.

Developed an AI-driven people counting system, leveraging computer vision

techniques to detect, track, and estimate pedestrian movement in diverse
environments.
Designed and implemented an adaptive background modeling system, using
Gaussian Mixture Models (GMM) to differentiate foreground objects from static
backgrounds.
Optimized object detection and tracking algorithms, integrating Kalman Filters
and the Hungarian Algorithm for accurate real-time multi-object tracking.

Page 6 of 8
Implemented morphological image processing techniques, reducing false
positives and enhancing the accuracy of object segmentation.
Developed k-Nearest Neighbors (k-NN) classification models to accurately
count people in high-density environments, mitigating occlusion challenges.
Integrated an FSM-based (Finite State Machine) system for directional
movement tracking, enabling multi-directional counting for advanced
pedestrian flow analysis.
Built a prototype for real-world testing, validating system performance across
low-light and variable crowd density conditions.
Applied machine learning models for crowd behavior prediction, enabling
proactive security monitoring and public space management.
Collaborated with interdisciplinary teams, combining expertise in AI, urban
mobility, and surveillance technologies to optimize deployment strategies.

Instituto Centro de Ensino Tecnologico

Statistical Analyst
janeiro de 2011 - novembro de 2013 (2 anos 11 meses)
Fortaleza e Região, Brasil

At Instituto CENTEC, I contributed to large-scale educational analytics

projects, focusing on data-driven policymaking and system development to
improve academic and social outcomes. Highlighted Projects & Contributions:

Designed and managed large-scale socio-economic research studies,

evaluating academic performance and social impact metrics.
Developed statistical models to analyze student success rates, supporting
government initiatives in education.
Created a dynamic data visualization platform using R and Java, enabling real-
time reporting for education policymakers.
Built automated data collection and analysis pipelines, integrating SQL
databases with Moodle's learning environment for tracking student
engagement and performance.
Provided insights for national and state-level education strategies, assisting
ministries and secretariats in decision-making processes.

Formação acadêmica
UNINTER Centro Universitário Internacional
Bachelor's degree, Psychoanalysis · (novembro de 2025)

Universidade Estadual do Ceará

Page 7 of 8
Master's degree, Computer science · (2014 - 2016)

Universidade Federal do Ceará

Bachelor's degree, Statistics · (2005 - 2010)

Page 8 of 8