Société Générale · Apr 2022 – Mar 2024
NAV justification model
A production model that helps explain movements in a fund’s Net Asset Value.
Senior Data Scientist
Productiondeployed and scheduled
K8scontainerised on Kubernetes
The challenge
In investment management, movements in Net Asset Value (NAV) need to be justified. Doing that reliably needed a model that ran as dependable production infrastructure, not a notebook.
How it works.
Decision-tree models
Machine-learning decision trees model the drivers behind NAV, keeping the logic inspectable.
Event-driven pipeline
RabbitMQ messaging connects the stages, with data in AWS S3 and PostgreSQL.
Orchestrated and containerised
Airflow schedules the workflows; Docker and Kubernetes provide scalable, repeatable deployment.
The outcome
A production-ready predictive model for NAV justification, built on a scalable, operationally efficient stack.
My role
- Developed the model and its production pipeline end to end
Stack
- Python
- Decision trees
- RabbitMQ
- AWS S3
- PostgreSQL
- Airflow
- Docker
- Kubernetes