Andrei Alexandru Stefan
MSc in AI @ Uniba. Full-Stack Developer. Italy.
About
I'm a Full-Stack Developer and Master's student in Artificial Intelligence at the University of Bari.
My work spans software engineering, machine learning, and AI research. I'm particularly interested in large language models, knowledge graphs, and building reliable software systems.
Here you'll find a selection of the projects I've worked on throughout my academic and professional career.
Latest work and writings
The most recent projects and articles.
Project Interpretable Image Retrieval
SLIDERS: Steering Image Retrieval with Sparse Visual Features
A patch-level image retrieval system in which sparse visual features become named controls for changing the ranking at query time.
Python, PyTorch, DINOv2, Sparse Autoencoders, FAISS, FastAPI, Vision-Language Models
Project Deep Learning
Replacing Qwen's Vision Transformer with ResNet-50
A feature-distillation pipeline for replacing Qwen's visual encoder with ResNet-50 while keeping the language model unchanged.
PyTorch, ResNet-50, Qwen-VL, Vision Transformers, CNN, Feature Distillation
Knowledge Graphs 11 min
The Bottleneck Wasn't the LLM: What Actually Improved Text-to-SPARQL
An ablation study on Text-to-SPARQL generation showed that reliable entity linking mattered far more than extra examples, schema hints or increasingly elaborate prompts.