Success stories



Very strict quality processes are enforced in the aeronautic industry. Non-Conformities observed during production trigger additional work and delays. They are a major blocker for production ramp-up. Non conformities are described through free text with technical abbreviated langages and codification.

D3S team developed a focused transformer-based AI to measure similarity between NC, allowing fast classification, nearest neighbors identification and hidden reoccurence pattern detection.


Data covered x00 000 descriptions of non conformities in several languages.

Our Tech Stack includes : PySPark, PyTorch, Palantir Foundry, Workshop, Hubble objects, Hugging Faces transformers, AWS, … and of course the NLP librairies from our lab ;


Huge manual effort is required in Quality and Operations teams to try and eradicate recurring non-quality. The clustering and identification of re-occurrencies require time consuming analysis almost impossible to perform with sufficient consistency. There have been multiple past initiatives to analyze these texts automatically through standard methodologies (RF or LR on tokenized words or n-grams, expert rules, etc) but none achieved the required robustness.

Objective of the project was to develop breakthrough digital capabilities to accelerate root cause analysis and eradication on the most recurring topics. Our advanced NLP algorithms (transformers) identify relevant patterns, detect automatically re-occurrencies and classify non conformities for further analysis. Results are pushed to a Single Source of Truth (SSOT), with a clear ontology and accessible to user’s through an interactive interface.


D3S team developed an innovative transformer-based AI clustering non conformities with >90% accuracy compared to former manual work.

Re-occurrences and main patterns of defects are now automatically identified. An integrated data set and cross-functional software is available to manage root cause analysis and eradication action plans.