Added screenshots, links to project page.

Pull Request with details

Initial visualization of the Valentine scores of datasets. Fixed notebook uploaded to Renku.

Added an interesting presentation of the Valentine project from TU Delft

Sketch from a blog post I wrote in February, on Trustable Integration. Seems relevant today ;-)

In a nutshell, here is what we want the initial prototype to do:

Take a user prompt from free text, as well as an uploaded file, then call an MCP service of the BFS i14y.admin.ch with this context, making sure to use the sessionID. The idea is that the MCP service will return a set of datasets based on the topics and needs described. The datasets are returned in a semantic ranking. The user chooses up to 3 datasets, and we calculate their structural compatibility, based on the metadata. Use the Valentine Python library to do this. Generate a report based on the user selection, with an indication of what needs to be merged or transformed in order that those selected datasets can be used together.

Over lunch & desert (pictured above 😋), we discussed how automation could support users of the platform: in the future, this could be to answer general questions via RAG, as well as to recommend datasets from the catalogue. Today, we want to focus on a specific need of data analysts: to estimate the degree of harmonization of a set of datasets, suggest compatibilities and pitfalls from a methodological perspective. This is something I can support with the Apertus LLM

In the morning we got demos of the platform and explored a couple of use cases, including this dashboard visualizing regional health indicators from an HSLU research project.