Ricardo Luís Bertolucci Filho

Mathematician | Data Scientist

Data Science • Machine Learning • AI • Mathematics

About Me

I hold a Master's degree in Mathematics with a specialization in Functional Analysis and Operator Theory. Currently working as a Data Scientist, combining rigorous mathematical foundations with practical data-driven solutions and modern AI systems.

My work sits at the intersection of machine learning, AI engineering, and applied mathematics. I design and build intelligent systems that go beyond traditional ML pipelines, including LLM-powered applications, multi-agent architectures with LangGraph, and RAG systems that combine retrieval and generation to deliver context-aware, reliable outputs. I use Claude (Anthropic) and GitHub Copilot as daily tools for development, code review, and AI-assisted engineering workflows.

I believe that the mathematical rigor behind optimization, linear algebra, and probability theory is what gives AI systems their interpretability and robustness. That perspective shapes how I approach every problem, from fine-tuning language models to designing stateful AI agents that reason, plan, and act autonomously.

Skills & Expertise

Programming

Python R Julia SQL Git LaTeX

Data Science

Machine Learning Deep Learning Statistical Analysis Data Visualization Predictive Modeling

Mathematics

Mathematical Analysis Differential Equations Linear Algebra Geometry and Topology Optimization

AI & LLMs

LangChain LangGraph RAG LLM Fine-tuning Prompt Engineering AI Agents Claude (Anthropic) GitHub Copilot

Featured Projects

Problem → Approach → Result

SupportMind AI — Evaluation-First Ticket Triage Assistant

In Development

Problem: support teams need to triage incoming tickets, classify issues, estimate priority, detect sensitive information, and generate reliable responses grounded in internal documentation.
Approach: building a production-style AI assistant with FastAPI, Pydantic, RAG, LLM-based classification, priority scoring, PII detection, structured logging, and an evaluation-first workflow for iterative improvement.
Result: currently in development as a portfolio project focused on LLM orchestration, retrieval quality, observability, safety checks, and product-oriented API design.

PythonFastAPIPydanticLLMsRAGVector DBEvaluationObservabilityDocker
Repository coming soon

Heart Disease Predictor

Problem: estimate heart disease risk from routine clinical variables in a way that is both predictive and interpretable.
Approach: built a PyTorch MLP with BatchNorm and Dropout, added SHAP-based explainability, and packaged the workflow with FastAPI, Streamlit, and Docker for interactive use.
Result: AUC-ROC ~0.92 Accuracy ~0.85 Recall ~0.87

PyTorchFastAPIStreamlitSHAPDocker
View on GitHub

Interactive ML Playground

Problem: machine learning models can be difficult to understand when their behavior is only presented through static metrics or code.
Approach: built an interactive Streamlit app featuring 7 classifiers, 8 datasets, decision boundaries, confidence maps, and comparative evaluation tools for visual exploration.
Result: 7 algorithms 8 datasets full metrics dashboard.

Streamlitscikit-learnPyTorchPlotly
View on GitHub

Additional Projects

Selected work beyond the featured highlights

Finance Data Science Project

Problem: analyze financial time series from multiple angles, including forecasting, classification, anomaly detection, and risk estimation.
Approach: combined LSTM forecasting, feedforward neural networks for direction classification, autoencoder-based anomaly detection, Monte Carlo simulation, SHAP explainability, and Value at Risk analysis.
Result: LSTM MAE 2.86 USD VaR 2.73% 25 anomalies flagged

PyTorchLSTMSHAPMonte Carlo
View on GitHub

Mathematics Self-Study Roadmap

Problem: self-learners often struggle to find a coherent long-term path through mathematics that balances rigor, progression, and self-study-friendly resources.
Approach: designed a bilingual roadmap that organizes mathematical learning into structured tracks, curated book recommendations, and a clear progression from foundational topics to advanced undergraduate and graduate-level material.
Result: created a practical educational resource that reflects curriculum design, technical communication, and the ability to structure complex knowledge in a clear and accessible way.

HTMLCSSEducationBilingualCurriculum Design
View on GitHub

Currently

Languages

English: Advanced

Portuguese: Native

Studying

LLMs from Scratch — Sebastian Raschka

Designing Machine Learning Systems — Chip Huyen

AI Engineering — Chip Huyen

Building Applications with AI Agents — Albada

Exploring

Multi-agent systems with LangGraph

Retrieval-Augmented Generation (RAG) pipelines

LLM evaluation & observability

Claude API & Anthropic tooling

GitHub Copilot for AI-assisted development

Contact

Reach out for collaborations, opportunities, or questions.