Alexandre Daniel

Facilities Management:

End-to-end support for the redesign of a cloud application to optimize maintenance operations planning

Definition of the application and software architecture (Airflow – Docker - Databricks – Pyspark – Azure Data Factory)

Definition of testing, release, monitoring, alerting and cyber security strategies

Refactoring of the application (according to clean code principles) Versioning (Azure DevOps)

Implementation of CI / CD

Orchestration with Airflow

Production launch in December 2023

 Combating Money Laundering and Terrorist Financing (AML-TF):

End-to-end support for the creation of a regulatory cloud application to identify at-risk customers

Definition of the application and software architecture (Snowflake – Databricks – Pyspark – Azure Data Factory)

Application development (according to clean code principles)

Versioning (GitLab)

Orchestration with Azure Data Factory

Monitoring (with MlFlow and Azure Data Factory)

Production launch in June 2022 and documentation

 Mail classifications:

Objective: to direct incoming emails in the CRM (salesforce) to the right people

Project scoping with customer service

Definition of performance objectives (90% reliability)

POC using Databricks and HuggingFace

Classification using a logistic regression and the CamemBert deeplearning model (fine tuning of the last layer): reliability achieved on 17 categories = 80%

 MlOps Architecture:

Definition and implementation of an MlOps architecture based on Databricks, Snowflake, MlFlow and Azure

Application architecture (experimental phase, deployment and monitoring)

Management of the implementation of a real-time fraud scoring model from scoping to production in June 2023.

 Basel3 Default Engine:

Cloud application to identify contracts in default of payment

Code refactoring (pep8 standard, evolution towards a service-oriented software architecture, clean code)

Development and development of new features (forbearance management, …)

Implementation of tests (unit, regression, …) and monitoring

Optimization of processing times (upstream filtering, caching, ..)

FinOps: Cost monitoring and optimization (choice of the most suitable VMs)

Security: static code analysis

Production launch in September 2022

Regulatory application audited and validated by KPMG

 SAS to Databricks Migration:

End-to-end support for the creation of an ETL application allowing the ingestion of SAS tables in sas7.bdat format into Snowflake via Databricks

Architecture (Snowflake – Databricks – Pyspark – Azure Data Factory)

Development, versioning, orchestration, monitoring, documentation

Note: the SAS table schema is completely deduced from its reading in Databricks. The data types defined in Snowflake by the application then allow a rereading by SAS.

Production launch in June 2023

 Library:

Design, development and production of a PySpark library to facilitate data exchange between Databricks and Snowflake

 Framework:

Design of a framework to quickly set up a python application based on PySpark and Snowflake

 Management

• Supervision of a team of 4 data engineers

• Animation with the Scrum method (sprint, agile rituals, …)

• Skills development of the team (code review, …)

• Backlog management: projects, evolution, maintenance, planning

• Cost monitoring and optimization (Databricks, Snowflake)

• Technical monitoring

 Document classification:

use of a deep convolutional neural network (InceptionV3) to distinguish 3 administrative documents (tax notice page1, 2&3, CPAM statement)

Results: 97% accuracy

 Detection of fraud in incapacity for work:

Project scoping, modeling (combination of two models: RandomForest and Gradient Boosted Tree), production launch (MLflow, Azure Databricks)

Definition of KPIs and monitoring of the algorithm in production (control chart)

Results: 72% accuracy, estimated indu of €7.6 million

 Creation of a ‘Feature Store’ (variable warehouse for data scientists):

Needs scoping, solution architecture

Implementation and deployment as a Python package with Pyspark and the ‘Feature Factory’ Framework from the Databricks lab.