AI software developer · BSc in Data Science & Artificial Intelligence, SUPSI
I build local-first AI systems, websites, and web apps. My focus is on retrieval-augmented generation (RAG), source-grounded answers, and turning practical problems into working software.
I completed my BSc at SUPSI in Switzerland in 2026. Alongside my AI work, I am a self-employed web and AI consultant for small businesses in Italy.
For my bachelor's thesis, hosted by Alwicom SA, I designed and evaluated a local RAG system over 44 clinical guidelines and 3,236 PDF pages.
- Built metadata-aware PDF ingestion, dense retrieval with guideline routing, cross-encoder reranking, and answers with document and page citations.
- Ran local inference with Qwen via Ollama on NVIDIA Jetson AGX Orin, using bge-m3 embeddings and PostgreSQL/pgvector.
- Retrieved the expected guideline in 98/99 benchmark questions. Evaluation included LLM judges, grounding analysis, ablations, and expert clinical review.
These are thesis benchmark results, not clinical validation. The portfolio explains the evaluation and its limitations.
Explore the thesis and pipeline walkthrough
| Project | What I built | Stack |
|---|---|---|
| wordcaps | Local word-by-word video captions, with speech-aware timing, editable transcripts, and output coverage checks. | Python, Whisper, ffmpeg |
| geocheck | A website audit CLI and library for search and AI discoverability, with evidence for each finding and zero runtime dependencies. | TypeScript, Node.js, GitHub Actions |
| Citation link prediction | A collaborative SUPSI project comparing tabular models and a graph convolutional network for scientific citation prediction. | Python, LightGBM, PyTorch, PyTorch Geometric |
| gym-tracker | An offline-capable gym log I use with friends, with workout history and progress tracking. | React, TypeScript, Vite, PWA |
TaxAI: co-founded and built a pre-launch MVP for Italian tax information, combining RAG with source-linked answers, document storage, and personalised fiscal deadlines. Built with Next.js, FastAPI, and PostgreSQL/pgvector. Currently paused.
Independent consulting: websites, booking flows, and web apps for small businesses. Public work includes Greenjoy Piscine and Emanuela Frassoldi, whose booking flow connects to Google Calendar through Netlify Functions.
I work under the trade name Plasma, often alongside fellow independent professional Matteo Martinelli. My focus is development, search visibility, and AI integrations.
- AI & data: Python, SQL, machine learning, NLP, RAG, LangChain, Ollama, and pgvector.
- Web & APIs: TypeScript, JavaScript, React, Next.js, Astro, FastAPI, Flask, and REST APIs.
- Delivery: PostgreSQL, Supabase, Docker, Git, GitHub Actions, Vercel, and Netlify.
For case studies and project context, visit my portfolio. For professional enquiries, connect on LinkedIn.