Volunteering / hackathon mentor, speaker & jury

AI for Impact

AI for Impact is the hackathon organised by La Mêlée during the Mêlée Numérique; its 2026 edition ran on 9 and 10 October at ENSEEIHT, on challenges set by industrial and public partners. I took part as technical mentor, speaker and jury member. My focus was how teams read their data before modelling it: what each variable means, its unit, missing values, the target, and what a metric says about the problem.

Daniel M'Bouyou with Sacha Smietana and Chahyn Ettaghi in front of the La Mêlée Numérique partners' wall
FIG. 01 · AI for Impact · La Mêlée NumériqueWith Sacha Smietana and Chahyn Ettaghi.
01

What I did

  • Talk: reading a dataset with SkrubOn a public CSV: TableReport to inspect columns, distinct values and empty cells; fuzzy_join to match inconsistent keys; TableVectorizer to encode mixed columns; then a simple baseline before any complex model.
  • Talk: Pydantic in five minutesValidating the fields a language model extracts from invoices against an explicit schema, so errors surface before they reach the application.
  • Mentoring and juryTechnical support for the teams during the event, and evaluation of their prototypes.
  • PreparationA technical handbook with one chapter per challenge and a foundations chapter: data leakage, train and test splits, baselines, metrics expressed in the target's unit, and what a missing value means.
Slide from my Skrub talk: 443 distinct job titles in a single CSV column, explored with the TableReport function
FIG. 02Skrub talk · slide 10. A public CSV with 443 distinct job titles: TableReport exposes the problem before any model is trained.
02

On stage

03

La Mêlée

La Mêlée is the Toulouse digital association that organises the Mêlée Numérique and this hackathon. Toulouse IA, a separate student association, took part in the organisation and held its first mini-hack at La Mêlée in September 2026.