I'm an engineering student at Télécom Paris (Institut Polytechnique de Paris), pursuing a double degree with the University of São Paulo (USP).

I'm interested in Artificial Intelligence, Machine Learning, Data Science, and Software Engineering, particularly at the intersection of mathematical foundations and practical implementations.

Python · LLM Agents · Docker · Information Retrieval · ongoing

A Python platform for building, evaluating and, eventually, optimizing tool-using LLM agents. Every run is bounded, traced and scored by an independent verifier, so claims about what an agent did can be checked.

• Coding environment: the agent fixes Python bugs inside a network-isolated Docker sandbox. A separate verifier runs public and hidden tests, and deliberately wrong "decoy" fixes check that the tests catch plausible mistakes. First pilot runs use Claude Haiku 5.5.
• Retrieval environment (in progress): a BM25 baseline with standard retrieval metrics on the BEIR SciFact benchmark, the starting point for studying how harness settings affect retrieval agents.

Python · Causal Inference · Optimization

A Python library for causal inference in structural causal models, capable of identifying causal effects or computing bounds when exact identification is not possible.

The library supports multiple identifiability criteria and optimization-based inference methods. I contributed to its development, documentation, testing, examples, and project website.

Python · Docker · Causal Inference

A modular framework for running and benchmarking causal inference algorithms in reproducible environments.

Integrates libraries such as DoWhy, bcause, LCN, and Autobounds. I worked on Docker-based orchestration, solver integration, inter-service communication, and error handling.

TypeScript · React · Kotlin · Docker

A full-stack application combining movie and cinema discovery with public transportation route planning.

Developed as a team project at USP, using React and TypeScript for the frontend and Kotlin for the backend. My contributions included backend testing and data processing.

Kotlin · Ktor · PostgreSQL

Backend for a recipe-sharing application with user accounts, recipe management, tags, and favorites.

Developed REST API functionality and worked on database integration, user authentication, recipe management, and automated tests.

Graph Machine Learning · Python · Cython

Contributed to the research and development of UGAP, a graph dimensionality reduction algorithm based on Personalized PageRank similarities.

My work involved algorithmic experimentation, benchmarking against established dimensionality reduction methods, investigating initialization strategies and hyperparameter sensitivity, and exploring alternative approaches to Personalized PageRank computation.

View merged Pull Request #599 →

• Languages: Python, C++, C, Kotlin, TypeScript, SQL
• Machine Learning & Data Science: PyTorch, scikit-learn, NumPy, Pandas, NLTK, spaCy
• Development & Tools: Git/GitHub, Docker, Linux, PostgreSQL, Jupyter, React, Ktor
• Interests: Deep Learning, Natural Language Processing, Causal Inference, Graph Machine Learning, Information Retrieval & RAG, LLM Agents & Evaluation, AI Observability, and Software Engineering

I'm always interested in opportunities to collaborate on interesting problems in machine learning, data science, and software development.

• GitHub: @DaviCoelh0
• LinkedIn: Davi Coelho