MSc AI · Università della Svizzera italiana (2026) I build ML/NLP systems end-to-end: from raw data through training, evaluation, and deployment as a working, documented, production-ready service.
🟢 Available for hire — ML Engineer · Data Engineer · NLP/CV Engineer · Switzerland & EU
I'm a fresh MSc AI graduate from USI Lugano with a full-stack software engineering background. I build complete ML systems; not just models, but pipelines, APIs and deployed services with monitoring. My thesis focused on few-shot visual learning for pest detection in precision agriculture and my broader project work spans LLM fine-tuning, RAG agents, NLP tooling and imitation-learning robotics policies with full MLOps pipelines.
| Project | What it shows | Stack |
|---|---|---|
| insurance_agent | RAG + tool-calling agent (calculator + glossary) over policy docs, served via FastAPI | LangChain · LangGraph · FastAPI · RAG |
| EagleSFT-Bilingual-Dataset-and-LLM-FineTuning | 536K-example EN↔RU SFT dataset built end-to-end + Qwen2.5 LoRA fine-tune | HuggingFace · LoRA · Qwen2.5 |
| DPO-Safety-Alignment-Phi3 | DPO/QLoRA safety alignment of Phi-3-mini on a single T4 GPU, with eval harness | DPO · QLoRA · PEFT · PyTorch |
| manipulation-policy-pipeline | Imitation-learning policy with full MLOps: FastAPI · Docker · Kubernetes · Prometheus/Grafana | MLOps · BC · Diffusion Policy |
| AI-Resume-Skill-Extractor | NLP tool scoring resume–JD match (TF-IDF + spaCy), packaged, tested, containerised, Streamlit UI | NLP · spaCy · Docker · CI/CD |
| aspilanta-fewshot-detection | MSc thesis — few-shot pest detection on yellow sticky traps in precision agriculture | Few-shot · CV · PyTorch |
