AI Scientist & Researcher · ACS-USP

Felippe Alves

Building AI systems that think in complex, uncertain worlds — from multi-agent architectures to Bayesian reasoning at scale.

Felippe Alves

About

I am a Postdoctoral Researcher at ACS-USP (Agentic Complex Systems — University of São Paulo), working at the intersection of AI research, complex systems, and multi-agent intelligence. My background spans a PhD in Physics focused on statistical mechanics, machine learning, and societal polarization modeling, followed by postdoctoral research at Aston University on epidemic control and optical communications. Today I design and study AI agents — their reasoning, collective behavior, ethics, and emergent dynamics.

Experience

2025 — now
AI Scientist & Researcher
ACS-USP · Agentic Complex Systems, University of São Paulo

Research on AI agents and multi-agent systems, focusing on emergent collective behavior, moral susceptibility of language models, ethical alignment, and agentic architectures grounded in complex systems theory.

LLM Agents Multi-Agent Systems AI Ethics Complex Systems
2024 — 2025
Senior Data Analyst
Fastshop · Data Science

Machine learning and model development for business processes in retail, including sales performance analysis, price tracking, and AI-driven product matching.

Python ML Databricks Spark
2022 — 2024
Postdoctoral Researcher
Aston University · Applied Mathematics & Data Science, Birmingham, UK

Two concurrent projects: (1) Optimal control of epidemic spread via SIR model on complex networks using Dynamic Message Passing, investigating conditions under which resource sharing is collectively beneficial. (2) Optimized QAM signal decoding for optical fibre communications via Generalized Belief Propagation, achieving up to 20% improvement over standard methods.

Julia Belief Propagation Optimal Control Complex Networks
2019 — 2020
Associated Researcher
GPOPAI-USP · in partnership with Social Science One & Facebook

Analyzed the demographics of hyperpartisan news sharing in Brazil, identifying polarization intensity and its sociodemographic distribution across Facebook communities using large-scale out-of-memory data pipelines.

Python SQL Social Networks Polarization

Research Themes

AI Agent Theory
Foundations of agentic behavior, planning, and autonomy in language model systems.
Multi-Agent Systems
Collective dynamics, emergence, and coordination in populations of AI agents.
LLM Ethics & Alignment
Moral susceptibility and ethical reasoning under adversarial or ambiguous conditions.
Bayesian Methods
Probabilistic inference, belief propagation, and uncertainty quantification.
Complex Systems
Statistical mechanics of interacting networks: opinion dynamics, polarization, epidemics.
Scientific AI
Agent-driven automation of the research pipeline — literature, hypothesis, experiment.

Education

2015 — 2020
Ph.D in Physics
University of São Paulo

Statistical mechanics of affective-ideological polarization. Developed an optimized ML algorithm enabling neural networks to assign trust/distrust to information sources, and discovered an entropic learning dynamics for neural networks.

2012 — 2015
M.Sc in Physics
University of São Paulo

Modeling of ideological polarization in the Brazilian Congress and its relation with public opinion, using Monte Carlo simulation of interacting neural network systems.

2008 — 2011
B.Sc in Physics
University of São Paulo

Publications

2025
arXiv preprint
Moral Susceptibility and Robustness under Persona Role-Play in Large Language Models
DB Costa, F Alves, R Vicente
2024
Journal of Physics: Complexity
The Futility of Being Selfish in Vaccine Distribution
F Alves, D Saad
2024
Nonlinear Optics and its Applications
Examining Nonlinear Behaviors in Optical Communication Systems using Gaussian Mixture Model
E Sedov, F Alves, D Saad, S Turitsyn
2021
arXiv preprint
Homo Entropicus, the Emotional Agent and Societies of Neural Networks
F Alves, N Caticha
2021
arXiv preprint
Frustration, Glassy Behavior and Dynamical Annealing in Societies of Neural Networks
F Alves, N Caticha
2021
Bulletin of the American Physical Society
Modeling Discrimination in Societies of Neural Networks
M Mercucci, O Citton, F Alves, N Caticha
2019
ESANN
Trust, Law and Ideology in a NN Agent Model of the US Appellate Courts
N Caticha, F Alves
2016
AIP Conference Proceedings
Sympatric Multiculturalism in Opinion Models
F Alves, N Caticha

Skills

Languages
Python Julia
ML & AI
LLM Agents Transformers Sklearn Deep Networks
Data
Pandas Spark Databricks SQL
Methods
Bayesian Statistics Belief Propagation Optimal Control Monte Carlo
Tools
Git Linux Jupyter
Human Languages
Portuguese English Spanish
Felippe Alves · São Paulo, Brazil ↑ top