Hassen T.

At BNP Paribas, I joined the DataLab team to support the integration and operationalization of Machine Learning and AI projects within critical banking systems. Leveraging my expertise in ML & AI infrastructure and system design, I focus on building secure, scalable, and intelligent applications—particularly in the context of agentic AI systems, within Domino Datalab platform.

A significant part of my contribution revolves around helping Data Scientists and System Engineers to design and build AI systems tailored for banking use cases, incorporating retrieval-augmented generation (RAG) pipelines and agentic workflows. These applications often require close coordination with security, infrastructure, and compliance teams, especially when handling sensitive internal data or deploying to IBM Cloud or on-premises infrastructure.

Key contributions within this role:

- Integrating LLMs seamlessly into the existing DataLab ecosystem, ensuring they serve business needs while adhering to internal governance

- Auditing, monitoring, and optimizing deployed AI systems for reliability and security

- Designing and deploying chat-based applications that interface with sensitive banking data via RAG systems and fine-tuned LLMs

- Evaluating LLMs (e.g., Mistral, LLaMA, GPT) for optimal use depending on internal use cases

- Helping internal teams define AI-enhanced developer workflows, including the use of internal chat-based assistants

- Contributing to the architecture of future agentic systems to enable intelligent multi-step task execution

- Optimizing prompts and system design for robustness, efficiency, and clarity of response

This role involves a thoughtful blend of AI strategy, MLOps discipline, and infrastructure engineering—with an emphasis on subtle but high-impact enhancements to both developer experience and end-user capabilities. My work has laid the groundwork for secure, compliant, and intelligent AI integration across the BNP Paribas ecosystem.

BPI France is a French public investment bank and financial institution that aims to support the growth

and the development of French businesses, particularly small and medium-sized enterprises (SMEs).

Within BPI, I joined the AI Lab to help them define and deploy AI & Machine Learning applications. With

my strong experience in Data and AWS services, I was able to build robust and performant ML

applications for model training, as well as predictions serving to end customers.

This was possible using the right AWS services, and the use of Jenkins and Terraform to reliably provision

those services.

My goal during this mission was to provide Data Scientists with a stable and performant Data and AI platform on the AWS Cloud, using reproducible and robust deployment tools.

Some key requirements and constraints of the AI Lab within BPI are:

- Collect and organize key datasets for analysis and model training

- Improve overall security of our applications, with respect to BPI policies, like the use of HSM

Instead of KMS, use AWS Inspector to spot security issues, restrict IAM policies to least privilege

- Be responsible of FinOps aspects within the IALab

- Systematic use of Terraform and Jenkins to build and provision the necessary AWS services

- Be aligned with overall BPI technical requirements, especially Security Auditing

- Create robust Data pipelines down to DynamoDB and ElasticSearch

- Dockerize the necessary applications for model training through AWS Batch on GPU instances

- Set up CPU & GPU instances for Data Scientists with the necessary tooling (Python/Conda,

Jupyterlab, CUDA, Auto-shutdown, monitoring and alerting)

AXA Global Parametrics provides businesses parametric insurance based on climate risks. This requires collecting and processing a wide range of data to create climate models, evaluate customers' risks and communicate those risks to our different applications and teams. To achieve this, we needed to create and automate data applications, as well as create an Event-Driven Architecture to communicate new processed data availability to the various services consuming them. This led to the following challenges:

- Find the right way to collect and store geospatial and time series data

- Automate predictions that were manually done by analysts and data scientists

- Create data solutions that follow DevOps & DataOps principles

- Build an events layer to communicate data results to services consuming them

- Update and maintain the various services consuming those data

Key Achievements:

- Create various data ETL pipelines to process climate indicators and store them in an appropriate way for quick retrieval (3 stages architecture)

- Work with data scientists to take their models into production

- Create an event layer for data consumption

- Create CI/CD pipelines to automate AXA Parametrics internal applications and services

- Design architectures and manage their implementation with tech team

- Refactor data science scripts to allow high performant computing for climate indicators