Eva Blot

Within existing projects (JupyterLab for modeling, Dataiku for production).

• Gathering and analysis of business needs

• Development, adaptation, and correction of scripts (Python)

• Writing detailed technical documentation for knowledge management (Confluence)

Gathering and analysis of business needs from project teams.

Development of projects on Dataiku / DSS (Pyspark scripts, SQL or integrated DSS modules - git versioning), automation scenarios, tested bundle in acceptance, preparation for deployment and documentation.

Development of projects in SQL, concerning "legacy" projects: modify or add features (new fields, data joining and enrichment…) to the existing, in application of new insurance standards.

User assistance and skill development:

- Assistance with script development and correction for Dataiku users

- Definition of good development and platform usage practices

- Analysis of error logs for problem resolution (Dataiku and Spark)

Predicting water pollution at the inlet of wastewater treatment plants to optimize electricity consumption. Solely responsible for my part, in a team of 3/4 people in full remote.

- Skill development on plant operation, technological watch on models adapted to different problems and choice of a solution by water quality indicators (neural networks (RNN), simple linear regressions or hybrid solution, to refine predictions)

- Use of Python (Jupyter notebooks for development and training, and Python scripts for inference and production deployment)

- Aggregation of code (and models) from other developers and refactoring to comply with production code standards

- Use of the Keras framework for neural networks

- Out-of-scope development: PySpark testing for code optimization on Spark

- Writing detailed documentation for client-side data scientists and managing their feedback

- Task management using Trello

Image recognition POC, for the recognition and classification of neckties (collars with badges for internal employees). Application objective: grant access rights based on colors. In a team of 3 people on-site.

- Needs analysis, technological watch on CNNs

- Implementation of a first algorithm developed by R&D teams, trained for human recognition. Second CNN algorithm for collar recognition and classification (color).

- Use of Python (scripts) and bash scripts (image labeling upstream)

- Use of the Tensorflow framework for the neural network part

Predicting Atos opportunities using historical contract and competitor data.

- Data cleaning and analysis. Use of NLP techniques (for textual data: contract descriptions): word dictionary, tf-idf, stop word removing...

- Use of a neural network for classifying these contracts

- Use of R (Jupyter notebooks) for the code

- Implementation of Kibana dashboards to visualize data and results

- Use of Streamsets for transferring results to Elastic Search

Centralizing, Organizing, Analyzing, and Visualizing HR data for the R&D department. Partly in a team of 3 and partly independently.

- Participation in the development of the HR data centralization and anonymization process. Choice of sources, storage servers, and data encryption. This part was carried out in a team of 3 (data architect/data scientist)

- Analyses on R&D department HR data using R on JupyterLab among others

- Implementation of Kibana dashboards for visualizing data and analysis results. Streamsets for transferring results to Elastic Search

- Production maintenance and development of new features in parallel

- Agility allowing for scope evolution

- Task management on Polarion

Implementation of several small R&D projects, in a direction oriented towards research, state-of-the-art, prototyping, technological watch. The goal being to test and evaluate algorithms and tool functionalities for future use in internal or external projects.

- Testing Facebook's Prophet algorithm for a time series problem (performed on Suez project data)

- PySpark testing for the Suez project

- Testing Kibana functionalities to optimize searches on Elastic Search (custom queries): HR data

- Testing Kibana functionalities for direct predictions within the tool (Timelions): HR data

- Refining opportunity predictions on Atos WinLoss data following the methodology and algorithms from research papers (e.g., score-driven threshold, for binary and multiclass imbalanced dataset). Production deployment within a project.

- Testing sentiment analysis algorithms on customer feedback (NLP).