Quentin Debaeke

As a permanent employee atAlten Technologies

◉ I contributed to the development of aComputer Visionsolution for tracking people via infrared sensors usingPythonandTensorflow.My goal was to annotate new 80x80 images (+200,000 images), implement the architecture, train and optimize new models (+100) so that they could be deployed on an embedded target while respecting hardware constraints (<10ms inference, 200KB memory):

🎖️Expertise inComputer Vision, Python,**Tensorflow** andYolo

🌟Acquisition and annotation of newdata

🌟Data versioning withDVCto improve traceability and consistency

🌟Implementation ofYoloarchitecture andDeep Learningmodel training

🌟Bayesian optimization to approach the best possible performance ofDeep Learningmodels

🌟Model versioning and performance tracking withMLflow

🌟Quantization withTensorflow Litefor implementing models on embedded targets (STM32 range)

🌟Performance testing of the embedded model

◉ Additionally, I was responsible for developing a PoC for electric arc detection usingMachine LearningwithPython,TensorflowandAutoML, allowing for the analysis of electrical signal cycles and the identification of electric arc presence.

🎖️Expertise inTime Series,PythonandAutoML

🌟Understanding, creation, and annotation by clustering of newdata

🌟Data cleaning and extraction of relevant features

🌟Implementation and training ofMachine Learningmodels usingAutoML

🌟Model versioning and performance tracking with MLflow

🌟Popularization and awareness of AI solution development

Data Scientistwithin the Learning & Development team of the Laboratory of Psychology and NeuroCognition (LPNC)

◉ I contributed to the development of astatisticalanalysis and visualization tool for researchers in the team to identify dysgraphia in children, enabling better diagnosis of the disorder. This resulted in +20 newstatisticalindices and a diagnostic tool used by occupational therapists ⤵︎

🎖️Expertise inMachine LearningandStatistics

🌟Knowledge ofstatisticalmodels necessary for identifying dysgraphia

🌟Development of a tool specification document

🌟Correction of already implementedstatisticalmodels

🌟Addition of missing newstatisticalmodels

🌟Creation of a graphical interface for visualizing plots generated by patients from a graphics tablet withPyQt

🌟Training ofMachine Learningmodels (Random Forest) with EEG data

🌟Implementation of theMachine Learningmodel in the tool for occupational therapists

🌟Automatic generation of reports and conclusions ofstatisticalresults

◉ I also participated in setting up EEG data acquisition protocols for trainingMachine Learningmodels

As a fixed-term contract employee atCapgemini DEMS

◉ I contributed to the development of anAIsolution for similarity search based on existing specifications in the railway sector to facilitate responses to tenders. I delivered an internal benchmark titled "Text Vectorization and Similarity Search" and severalNLPmodels to meet the needs ⤵︎

🎖️Expertise inNLP,Python,TensorflowandBERT

🌟Technological watch of NLP approaches at the time for similarity search (Machine LearningandDeep Learning): TF-IDF, SVD, Doc2Vec,BERT(LLM)

🌟Preparation, cleaning, and creation of datasets from open-source and internal data in French and English

🌟Data preprocessing and analysis: stop words, generic words, lemmatization, n-grams, stemming

🌟Data annotation using hierarchical clustering (CAH)

🌟Data vectorization

🌟Implementation and fine-tuning of pre-trainedDeep Learningmodels (LLM)

🌟Analysis of difficulties encountered and potential approaches

◉ I assisted in the development of educational materials for learning programming,statistics, andAI(Machine Learning,Deep Learning, andReinforcement Learning) in the form of Jupyter Notebooks