Data Scientist | M.Sc. in Data Science @ Politecnico di Torino | B.Sc. in Mathematics | Ex-Data Analyst @ MCI (1+ year)
| Languages & Development | Python | C | SQL | Git | Streamlit | Gradio |
| Machine & Deep Learning | PyTorch | Hugging Face | Scikit-Learn | OpenCV |
| LLMs & Generative AI | LangChain | LangGraph | LlamaIndex | Ollama |
| Data Analytics & Visualization | Power BI | Excel | Pandas | NumPy | Matplotlib | Seaborn |
| Data Management & Big Data | PySpark | Hadoop | SQL Server | MongoDB | ChromaDB |
Multi-Agent Historical Fact-Checking RAG (In Development)
Python | LangChain | LangGraph | LlamaIndex | ChromaDB | DeepEval
- Objective: Build a fact-checking system that can verify historical claims using RAG.
- Architecture: Cyclic LangGraph workflow with specialized agents (Router, Librarian, Historian, Judge) over a ChromaDB + LlamaIndex retrieval stack with Nomic embeddings.
PyTorch | YOLO | ResNet | OpenCV | NumPy | Pandas
- Objective: Texture-less 3D object pose estimation targeting robotics and autonomous manipulation.
- Architecture: YOLOv10m detection combined with a ResNet18 heatmap regression network and multi-stage RGB-D cross-fusion.
- Result: 92.4% mean ADD accuracy on LINEMOD, outperforming the RGB-only baseline by +7.9%.
PyTorch | Torchaudio | Hugging Face (Transformers, PEFT) | Gradio | HTML/CSS
- Objective: High-fidelity speech emotion recognition evaluated on the complex ESD and IEMOCAP speech corpuses.
- Architecture: Voxtral-mini-3B audio-language model fine-tuned with LoRA and DoRA PEFT on ESD & IEMOCAP.
- Result: 0.84 macro-F1 in-domain and 0.63 zero-shot cross-domain generalization.
PyTorch | Hugging Face | RoBERTa | Adapters
- Objective: Detecting sarcasm, irony, and sentiment under severe dialectal and domain shifts.
- Architecture: RoBERTa-Large ensemble with Mixture-of-Adapters (MoA) and a Tensor-of-Cues (ToC) prompt-tuning mechanism.
- Result: +16% cross-variety sarcasm classification and +18% zero-shot dialect adaptation.