Skip to content
View pinthoz's full-sized avatar

Highlights

  • Pro

Block or report pinthoz

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
pinthoz/README.md

Hi there, I'm Ana Pinto 👋

I did my BSc in Artificial Intelligence and Data Science at FCUP/FEUP, University of Porto, and I am now finishing an MSc in Data Science.

The two halves suit each other. The degree gave me the AI side, how models are built and what happens inside them, and the master's gave me the data side, statistics, experimental design and the discipline of producing a result you can actually defend. I have ended up working at the seam between them, on whether a model's output can be explained and trusted rather than only measured.

That is what my MSc thesis is about: interpretability of large language models. I build tools that show what a Transformer attends to, and use those signals to detect and audit social bias in BERT and GPT-2.

Outside the thesis I build machine learning systems end to end, from messy data to something a person can actually open and use. I have worked on computer vision, robotics, audio, retrieval-augmented generation, reinforcement learning, tabular prediction, multi-agent systems and a fair number of web and desktop apps. The common thread is curiosity about why a model does what it does.

My Research Interests:

  • Interpretability and mechanistic analysis of Transformers
  • Explainable AI (XAI)
  • Bias detection and fairness auditing in language models
  • Faithfulness evaluation of explanations
  • Human-in-the-loop and human/AI auditing workflows

I am open to:

  • any collaboration work,
  • machine learning projects (development or research),
  • research internships and junior positions, and
  • writing or reviewing on interpretability, XAI and evaluation.

Skills:

Languages:

Python  Java  JavaScript  Kotlin  C#  C  C++  R  SQL  LaTeX  Markdown

ML/DL

PyTorch  Tensorflow  Hugging Face  scikit-learn  spaCy  OpenCV  NumPy  Pandas

Visualisation

Plotly  D3.js  Matplotlib  Seaborn  Shiny

Web and Apps

FastAPI  Flask  Django  Streamlit  React  Next.js  HTML5  CSS3  .NET

Database

MySQL  SQLite

Tools and Technologies

Git  Docker  Google Cloud  GitHub Actions  Linux  Figma  Photoshop

IDEs

Jupyter Notebook  Google Colab  Visual Studio Code  PyCharm

Some things I have built:

Interpretability and fairness

  • Attention Atlas  |  interactive platform for exploring attention in BERT and GPT-2, with a full bias-detection pipeline on top. Live demo
  • detecting-genre-bias-tweets  |  how the demographics of annotators shape the labels they give, on the EXIST 2025 dataset, with clustering, association rules and a recommendation framework
  • xai-ruralfires  |  explainability methods applied to rural fire risk

Language models and retrieval

  • benfica-rag  |  retrieval-augmented comparison of the electoral programmes of the two candidates in the 2025 Benfica elections, side by side, with page-level citations, in Portuguese and English
  • claude-notebook-skill  |  notebook generation for academic-grade data science work

Vision, audio and robotics

  • analogic-watch-detector  |  reads the time from photos of analogue clocks, YOLOv8/YOLOv11 plus hand-angle geometry
  • TabSynth  |  turns an audio recording into playable guitar tablature: pitch tracking, note segmentation and fretboard assignment
  • robotics-plane-detection-3d-mapping  |  RANSAC plane detection to improve precision in 3D mapping, on point clouds from a Webots simulation

Systems, apps and agents

  • hospital-backend  |  REST API for a hospital system, patients, admissions, progress notes and media, in Flask on Google App Engine
  • regybox-strava-sync  |  writes the day's CrossFit workout into the matching Strava activity, unattended, on GitHub Actions
  • sports-extension  |  floating always-on-top bar with live scores, in C#
  • multi-agent-system-atc  |  decentralised multi-agent system for air traffic control
  • bloxorz-game  |  puzzle game with solvers in BFS, DFS, IDDFS, Greedy and A*

Connect with me:

Ana's GitHub profile summary

Most used languages Most committed languages

Pinned Loading

  1. bloxorz-game bloxorz-game Public

    Python/Pygame implementation of the Bloxorz block-rolling puzzle with 6 levels, plus an AI solver using BFS, DFS, IDDFS, Greedy and A* (Manhattan/Euclidean/Chebyshev heuristics). Text and graphical…

    Python

  2. dara-game dara-game Public

    Web version of Dara, the two-player West African strategy board game. Includes user accounts, a ranking system and a browser-based board. Built with vanilla HTML, CSS and JavaScript.

    JavaScript

  3. frozen-lake-rl frozen-lake-rl Public

    Reinforcement learning on a customised Frozen Lake environment: custom movement and reward wrappers over OpenAI Gym, with training/testing scripts and a Jupyter notebook. Built with Stable Baselines.

    Jupyter Notebook

  4. analogic-watch-detector analogic-watch-detector Public

    Computer-vision system that reads the time from analog clock images. Uses YOLOv8/YOLOv11 to detect the clock face and hands, then computes the time. Includes a Tkinter GUI, a Gradio demo and real-t…

    Jupyter Notebook

  5. attention-atlas attention-atlas Public

    This repository hosts the code, experiments, and prototype developed for my Master’s thesis: Interpretable Large Language Models through Attention Mechanism Visualization. It provides tools for exp…

    Python

  6. sports-extension sports-extension Public

    Follow live scores without juggling browser tabs. A floating, always-visible bar streams your starred FlashScore games across football, tennis, basketball and more. .NET 9 · WPF · WebView2 · Chromi…

    C# 1