ML-powered delinquency nowcasting and economic indicator dashboards for institutional investors.
The challenge
Institutional investors tracking truck loan health had no purpose-built index. Hudson Insights needed a production system that ingested multiple external data sources, calculated a delinquency index with meaningful segmentation, and surfaced actionable signals before full reporting cycles closed.
How we built it
I built the data pipeline as a Databricks medallion architecture: raw ingestion in the bronze layer, normalization and transformation in silver, analytics and index calculation in gold, and ML nowcasting models in a dedicated tier. A FastAPI server exposes the index data to a React dashboard with role-based views for analysts and investors.
What shipped
HITLDI gives institutional stakeholders an early-warning delinquency index backed by diesel prices, unemployment, GDP, and freight volume data — updated on the pipeline schedule and exportable in CSV, Excel, and PDF for integration into external reporting workflows.
Outcomes
- Built a full medallion data pipeline (bronze → silver → gold) in Databricks, ingesting diesel prices, FRED economic indicators, and Cass Freight Index for index calculation.
- Delivered ML-powered nowcasting models for early delinquency prediction alongside a FastAPI backend with role-based access for analysts and investors.
- Created an interactive React dashboard with exportable reports (CSV, Excel, PDF) for institutional stakeholders.