This is a submission for the Hacktoberfest Weekend Challenge: Build for a Friend
What I Built
I built a while set of resources to link together job and skills referentials :
👉 I needed these links as an interoperable dataset to help my students from IAE MIAGE (UNC - University of New-Caledonia).
They take part to the HackAVP a hackathon that takes place here un New-Caledonia, dedicated to job offers and interoperability, Open Data, and public Open Innovation:
By linking these two referentials, it makes it possible to enrich and better understand skills of a job offer...so the apply process should benefit from a quality bump thanks to enriched ESCO semantic (which is much richer than ROME one from my point of view).
📑 Demo
I did not yet have time to make a video on this schedule as my Kaggle notebooks took hours of GPU to produce the data.
Here is the notebook that computes the links :
... which produces and shares data into a dataset :
... then produced a dedicated analysis Notebook :
Code
To achieve this work, I needed a lot of data, here it is :
- github.com/adriens/odata-esco which takes ESCO csv exports and builds a duckdb database on top of it to make data as smart and usable as possible
- Kaggle ESCO v1.2.1 dataset : the previous duckdb ESCO database as a Kaggle dataset
- Kaggle loader ROME v4 base DuckDB to prepare the Rome v4 dataset
- Kaggle Rome v4 Dataset dataset
- Rapprochement ROME v4 ESCO 1.2.1 : The mapping Notebook
- ROME ESCO mapping Dataset
- ROME ESCO mapping analysis : some analysis on the resulting Kaggle dataset
🧰 How I Built It
I built everything on Kaggle, on my GPU free plan:
I only used Open models on Kaggle :
- EmbeddingGemma
- gemma-4-e4b-it
- gemma-4-12b-it
- gemma-4-31b-it-qat-q4_0-gguf
- hf-model/BAAI/bge-reranker-v2-m3
🤗 Why Does Open Innovation Matter?
Open Innovation matters for a living democracy, to spread knowledge, skills and efficience of public services.
This is why I spend so much effort on this topic, like reading books, producing prototypes and teaching it to students to show that it works for real, in real life and has a tremendous impact on our society and freedome, especially when linked to Open Source software and Open Data.
And if the result or data does not really fit your needs or don't really match your expectations... you can do your own.
At last, it brings transparency... which is central by many concerns. This is why I wanted to link french referential with the european one.
🏆 Prize Categories
- Best Use of TabPFN
- Best Use of Gemma
📷 Screenshots
When Gemma disagrees with TabPFN, be carefu
Precision of the model's first choice, split by whether Gemma E4B picked the same ESCO occupation. The black tick is the model alone, all cases together.







Top comments (0)