Backend software engineer based in Geneva. I've spent the last 5+ years at CERN building and maintaining scientific information platforms — primarily CDS and its migration to InvenioRDM, an open-source research data management platform used by institutions worldwide.
I work mostly in Python, with a focus on REST APIs, search infrastructure, and data workflows.
Backend — Python, Flask, PostgreSQL, OpenSearch/Elasticsearch, Celery, REST APIs
Frontend — React, TypeScript
Infrastructure — Docker, Kubernetes/OpenShift, CI/CD
Domain — research data, open science, metadata standards, scientific repositories
A metadata assistant for scientific papers built to learn LLM tool calling and agentic workflows hands-on. Uses the Anthropic API and queries the INSPIRE HEP database. The model decides which tools to call and in what order — search, citation lookup, metadata fetch — without that sequence being scripted.
An MCP (Model Context Protocol) server that exposes InvenioRDM record management as tools an LLM can call. Lets an AI assistant create drafts, set metadata, upload files, and publish records on a research repository through natural language.
A retrieval-augmented generation (RAG) system for question-answering over scientific paper abstracts. Built in stages to understand each component: hybrid search (dense vector + BM25 fused with RRF), cross-encoder reranking with a relevance threshold, contextual RAG enrichment, and grounded generation with Claude. Indexes real papers from the arXiv API.
Two LLM modules for processing scientific paper abstracts: a metadata extractor that returns reliable structured JSON (title, authors, identifiers) from clean or messy input, and a classifier that assigns arXiv-style categories using both zero-shot and few-shot prompting — with a side-by-side comparison to show where each approach wins.
Most of my open-source work is through contributions to the InvenioRDM ecosystem — the framework behind Zenodo, CDS, and a number of institutional repositories.