I’m Mario, an AI engineer from Basel. I build AI solutions that make everyday work easier, from privacy‑first assistants and knowledge systems to computer vision.
Currently AI Solution Engineer at Megasol Energie AG, after completing my BSc in Artificial Intelligence (Business AI) at FHNW. My focus is turning ideas into robust, privacy‑first products, with models hosted in Switzerland whenever data must stay in the country. Outside of work I’m an endurance athlete: Ironman finisher and Swisspeaks 100‑mile finisher.
Moved into a permanent role after my internship: I find where AI can make Megasol’s processes faster and smarter, and build the solutions that make it happen.
Built and productized Protocolito, a privacy‑first AI meeting assistant. Originally a practice project with FHNW, continued independently, and brought to production at pixon.
Supported the indexing team of the open‑access publisher in Basel, working with JIRA, Redmine and Microsoft Office.
Privacy‑first AI meeting assistant as an Electron desktop app. Records, transcribes and summarizes meetings and extracts tasks, speakers and action items. All LLMs run in Switzerland. Started as an FHNW practice project, now in production at pixon engineering.
Computer‑vision thesis: image preprocessing, object detection, an annotation pipeline and ML‑based model optimization for a real industrial use case.
Technicians shouldn’t have to type reports. They record a voice note, Gemini extracts the hard facts (materials, hours) and the app maps them onto the company’s existing PDF templates. Works offline and syncs when back online.
Project module with two use cases: automatic meeting summaries (the first version of Protocolito, built on the open‑source project Meetily) and a web app that drafts feasibility studies with AI, using the company’s key sources as knowledge base. With Umut Can and Dewa Audric.
AI chatbot for eBay customer inquiries with a FastAPI backend and React frontend. Uses RAG on eBay help pages for precise, grounded answers.
Security system on a Raspberry Pi with a Camera Module 3 Wide. A trained model recognizes a set of saved faces and triggers an alarm when an unknown person appears. Built for the university module “Machine Learning”.
I post about what I’m building and learning. Endorsements, recommendations and the full history live there.
Visit my profile ↗Bachelor of Science (FH), Artificial Intelligence — Business AI
Exchange semester, Madrid
Berufsmaturität + EFZ, Basel‑Landschaft
Always happy to talk about AI, computer vision and interesting ideas. Reach out any time.