FiscalNote streamlines ML and NLP legislative-analytics model deployment on Databricks, tripling release cadence

Overview

FiscalNote, a legislative and regulatory data analytics company, used Databricks MLflow and Model Serving to automate tracking, deployment and no-disruption updates of its machine learning and NLP models, which perform tasks like legislative outcome sentiment analysis and binary classification of how a congressperson will vote. The change let FiscalNote deploy ML models 3x faster and increased the number of models it ships annually compared to before Databricks.

The challenge

FiscalNote's data teams were hampered by cumbersome, fragmented processes to stitch together components essential for AI model deployment such as tracking artifacts and builds, which limited their cadence for deploying and updating AI models to around once a year, required extensive custom coding to avoid disrupting existing models, and suffered from a lack of discoverability of essential data assets.

The solution

FiscalNote adopted Databricks MLflow to manage and deploy ML models, reducing time spent tracking artifacts, model versions and notebooks, and layered on Databricks Model Serving to deploy, manage and monitor models without needing to build APIs or plan no-disruption deployments manually, supporting ETL pipelines, NLP summarization and sentiment analysis, and binary classification workflows.

Reported business value

FiscalNote reduced bottlenecks with their data platform teams and can now deploy 3x the number of models annually compared to their output before Databricks, reducing time to market and allowing employees to be more productive and effective.