At Teverra, I was actively engaged in the development of data-driven subsurface monitoring solutions, from the initial stages of idea conception and value proposition, and proposals writing, to executing proof-of-concept demonstrations via data exploration, workflow development, software design and implementation, and testing within commercial project. In addition, I was the sole data scientist in a team of less than 10 geoscientists and engineers, and I supported different projects with data manipulation and exploration, code review and update and refining project task description and timelines.• Received 2 awards (over 1million $ each) from the US Department of Energy for proposals introducing physics-informed machine learning based technology, that enables real-time geomechanics modelling of subsurface formations during well drilling operations. • Designed and validated an end-to-end data analytics and machine learning pipeline transforming raw drilling dynamics sensor data, namely drilling vibrations and electronic drilling recorder data, into profiles of geomechanical rock properties.• Automated the data-analytics workflow by developing a python software tool able to seamlessly parse, normalize, pre-process, consolidate datasets of different temporal and spatial resolution into one unified dataset, generate data features and train and evaluate ensemble and neural network prediction models.• Validated the developed workflow and software within a commercial drilling project.• Regularly prepared and delivered powerpoint presentations and progress reports to communicate project progress updates to clients and funding agencies on a regular basis.• Conducted reviews of open-source code, adeptly rectifying errors and updating codes to align with our requirements, within geothermal and carbon storage projects. Reviewed, edited, and contributed to proposals, particularly those presenting data processing and ML components.

Following my relocation from Canada to the US, I held the postdoctoral position in the Neuro-Engineering department. I worked in the context of Alzheimer's Disease research. I conducted the literature review, designed and implemented behavior experiments to assess memory and learning functions in mice, acquired mice behavior data with Ethovision video tracking software, and conducted data analysis to validate experiments and build behavior baseline. In addition, I was involved in advising neuro-engineering students on their data science projects at Rice University.

I started my journey at TR labs as an intern and transitioned to a full-time role within four months following the delivery of a legal clause classification model that contributed to ongoing features development within a legal document analytics product. • Developed and evaluated end-to-end text processing pipelines for inferring content similarity across contract documents while leveraging large language models. • Worked toward adding features to Westlaw, a prominent legal search product by refining the categorization of legal cases and identifying relationships between them. • Built end-to-end predictive text processing pipelines, using embedding models, ensemble and neural networks models. • Developed expertise in leveraging AWS services for machine learning development.• Worked closely with legal subject matter experts, SMEs, to define and assign labeling and data annotation tasks that support the ongoing supervised machine learning developments, conduct error analysis and discussions aimed to bridge the legal domain knowledge gap, with the objective to refine the direction of the data science and software developments.

During my time at the University of Waterloo, Aquatic Informatics Inc. reached out to discuss my published research presenting an association rules modeling approach and a novel detection metric for identifying leaks in water distribution pipelines. They then engaged me as a consultant to explore the feasibility of identifying overflows in sewer pipelines. My work entailed the following:• Consolidated heterogeneous datasets into one unified dataset, and evaluated approaches for dealing with missing datapoints; withdrawing associated rows or imputing missing values through linear and non-linear interpolations.• Conducted multivariate correlation analysis and visualization to infer the combined effects of variables, such as the precipitation level, the water level, the temperature, and pump operation status on sewer overflows’ events.• Evaluated the performance of regression and classification techniques in predicting the occurrence of sewer overflows. Results were not conclusive and correlations were not statistically significant, mainly due to incomplete datasets resulting in high uncertainty.• Provided recommendations to complete the datasets addressing the large amount of missing data points, and involve more operational data to complement the existing datasets and explain the data variability.

After 8 months of job search in Canada upon relocating from France to Canada at the end of 2016, I joined the SDIC lab to support masters and PhD students with data collection, signal processing and machine learning, and code development, as well as conduct a R&D project funded by Pattern Discovery Inc aimed at developing a data mining solution tailored for monitoring water distribution pipelines and detecting small leaks using acoustic sensor data. • Developed a novel association rules mining approach and a novel leak detection metric processing acoustic measurements acquired from sensors mounted on the edge of fire hydrants .• Tested the developed workflow on real data acquired in the field. The prediction accuracy obtained in the lab proved to be significantly higher than in the field, due to the high-level noise existing in the field data and the lack of necessary operational data explaining the source of large transients. A high-impact paper resulted.• Developed and validated a data-driven approach combining time-frequency analysis and machine learning for the monitoring of gearbox systems, using the accelerometer data collected at the gearbox of the Terminal Train at the Toronto Pearson Int. Airport. A high-impact journal paper resulted.• Developed an unsupervided data processing method to detect leak events in water pipelines, by extracting discriminatory features from acoustic sensor data using time-series decomposition. A journal paper resulted.

This is an intensive fellowship, aims at identifying a business problem solvable with deep learning, developing a minimum viable product, and pitching the problem and the solution. • Proposed enhancing user experience on hotel booking websites, by offering time-based analysis of hotels reviews, including time-specific sentiments and opinions on various hotel features.• Developed a proof-of-concept through webscraping Tripadvisor.com, creating a software tool that transforms raw data into time-based insights allowing users to compare hotel performance over time.

During my postdoctoral position at the University of Waterloo, I joined IMS to apply my signal processing expertise in enhancing the assessment of driver behavior using telematics data. • Demonstrated the feasibility to detect and identify different types of abnormal risky driving behaviors (tailgating, jerks, etc.) using driver smartphone sensor data (gyroscope, acceleration, etc.) through time-frequency decomposition of smartphone sensor signals.• Developed an unsupervised algorithm that identifies driver primary and secondary locations of interest (home, school, coffee shop, etc.), using their historical driving geospatial data.

I joined the Signals, Systems and Physics team following my PhD to develop novel time-series decomposition techniques tailored for non-stationary signals. This work involved theoretical and analytical studies, coupled with simulation, code development within Matlab and testing on audio signals.• Developed a novel non-stationary, non-parametric time-series decomposition technique, called sliding singular spectrum analysis, and published the method in a top signal processing journal.• Compared the performance of the novel method with advanced time-series analysis techniques, such as the empirical mode decomposition and the synchrosqueezing transform.• Collaborated with the ASTRES toolbox developers to implement the new module/technique, and conduct simulation comparative studies.

During my internship at Liban Cables, I focused on enhancing the control programs for cable manufacturing machines using Ladder Logic Programming. I began by understanding the existing manufacturing processes and automation systems. By collaborating closely with engineers and technicians, I identified opportunities for improvements. I then proposed and implemented these enhancements through Ladder Logic, tested the changes, and documented the outcomes. This experience allowed me to contribute to optimizing the efficiency of the manufacturing process.

I delved into understanding the various control and automation processes integral to the production of cement and derivates. My role involved analyzing and gaining insights into the systems and workflows used to optimize production efficiency, ensure quality control and the monitoring and control of gas emissions. This experience provided me a direct exposure to SCADA systems, its application in a manufacturing environment and role in mitigating environmental risks.