PhD, Physics
Centro Brasileiro de Pesquisas Físicas (CBPF). Research in non-equilibrium thermodynamics and statistical mechanics.
Research & Writing
Non-equilibrium statistical mechanics by training; retrieval-augmented generation and semantic networks by current obsession.
Research focus
My doctoral work asks a deceptively simple question: what does "temperature" mean for a particle driven by a reservoir that is not thermal?
Using van Kampen's Langevin methodology, I studied particles under dichotomous (bimodal) and Gaussian coloured noise, derived their thermostatistics analytically and numerically, and compared alternative definitions of effective temperature. Along the way I examined fluctuation relations, the near-equivalence between external and internal reservoirs through large deviation analysis, and turbulent time series, with applications reaching into biology and finance.
The same instincts now go into AI work: treating a RAG pipeline as a graph with measurable structure, quantifying how much semantic space a knowledge base occupies, and asking what an anomaly detector's "normal" actually is.
Publications
Google Scholar →- Effective temperatures for single particle system under dichotomous noise João R. Medeiros, Sílvio M. Duarte Queirós · Journal of Statistical Mechanics: Theory and Experiment, 2021 doi:10.1088/1742-5468/ac014e · arXiv:2105.01185
- A large deviation analysis on the near-equivalence between external and internal reservoirs João R. Medeiros, Sílvio M. Duarte Queirós · Physica A: Statistical Mechanics and its Applications, 2016 doi:10.1016/j.physa.2016.01.029
- Thermostatistics of a damped bimodal particle João R. Medeiros, Sílvio M. Duarte Queirós · Physical Review E 92, 062145, 2015 doi:10.1103/PhysRevE.92.062145
- Temperaturas Efetivas em Sistemas Atérmicos Simples (PhD thesis) Centro Brasileiro de Pesquisas Físicas, 2024 · Advisor: Sílvio M. Duarte Queirós PDF (CBPF Index, Portuguese)
Education
MSc, Statistical Physics
CBPF. Thermostatistics of dynamical systems under exotic reservoirs, including Gaussian coloured noise and dichotomous reservoirs, combining analytical results with computational methods for out-of-equilibrium steady states.
BSc, Physics
Pontifícia Universidade Católica do Rio de Janeiro (PUC-Rio).
Writing
Medium →-
Retrieval Augmented Genesis — Holy Texts Semantics Analysis
Medium · October 2024
Explores the knowledge base behind RAGenesis. Introduces semantic hypervolume, measured through convex hulls of embeddings, as a metric of theme diversity; defines the intertext similarity fraction and intertext consistency to compare whole texts rather than chunk pairs; and shows how two embedding models disagree on the relationships between the Torah, the New Testament and the Quran.
-
Retrieval Augmented Genesis
Medium · October 2024
The thinking behind RAGenesis. Argues that every RAG application carries an implicit graph structure created by its embedding model, knowledge base and similarity metric, names it the Semantic Similarity Network, and shows how to use it for transparency and retrieval optimisation.