Neural Network Recommendation Engine
FEATUREDA two-tower recommendation engine deployed to 1.6 million users on the VT Markets app, driving cross-selling and product adoption among traders through personalised instrument suggestions.
PythonPyTorchPostgreSQLBigQueryBraze
Overview
Built during a data analyst internship at VT Markets, this recommendation engine personalises trading instrument content for 1.6 million users across the platform.
What it does
- Two-tower neural network architecture matching user behaviour to relevant trading instruments
- PostgreSQL and BigQuery ETL pipelines syncing user signals into Braze for downstream delivery
- Automated review outreach scaled across the full user base
- Internal LLM tooling enabling non-technical teams to run natural language queries on live data
The lesson
Recommendation systems in fintech have a cold-start problem that’s harder than most — new users have no trading history, and wrong recommendations erode trust fast. The ETL design mattered as much as the model.