ASO-Temba
CompletedExtensions to a Mamba-based model for temporal action detection in long videos, improving performance by up to 3.48 mAP points and enabling zero-lookahead streaming inference.
A comprehensive collection of my work in AI, machine learning, and software development. Projects span machine learning, natural language processing, and software development.
Extensions to a Mamba-based model for temporal action detection in long videos, improving performance by up to 3.48 mAP points and enabling zero-lookahead streaming inference.
A benchmark of six embedding families across 17 language varieties. Character TF-IDF outperformed pretrained Transformers in recovering linguistic relationships (Spearman ρ = 0.905).
A conversational AI prototype that helps students explore university courses in Italy through voice-guided interviews and personalized recommendations. Built for the innovAIte 2025 hackathon using a synthetic university dataset.
A multi-view computer vision pipeline that reconstructs 3D scenes from 2D images through camera calibration, feature matching, pose estimation, and triangulation, with interactive point cloud visualizations.
Bachelor's thesis exploring novel heuristics for efficient and accurate k-means clustering, including a proposed extended-Hartigan algorithm and its empirical evaluation.
A local, AI-powered tool that analyzes any public GitHub repository and turns it into clear, structured documentation, including concise summaries and visual overviews.
An end-to-end computer vision and Transformer pipeline for recognizing and translating Sumerian cuneiform, achieving 0.775 mean semantic similarity despite limited training data.
Statistical analysis and automated answer recognition system using Machine Learning (PCA and SVM-based classifiers). Developed with Hephaestus Applied AI Association.
Simulation study exploring the escape of a particle from an intracellular region using Brownian motion dynamics and statistical analysis.
Puzzle solver using genetic algorithms and island models to tackle complex constraint satisfaction problems through evolutionary programming.
Optimized JavaScript Brainfuck interpreter with loop unrolling, precalculated jumps, and operator compression for faster execution.
I'm always interested in new challenges and collaborations. Let's build something together.