Back office ops · Production

Databricks helps FiscalNote deploy ML models 3x faster

The problem

FiscalNote's data teams had to manually stitch together disparate components for AI model deployment, tracking artifacts and builds, which limited their deployment cadence to about once a year and stretched each deployment phase over several weeks. Custom coding was required to avoid disrupting existing models during updates, and poor discoverability of data assets made it hard for data scientists to find and use the data they needed.

Workflow diagram · grounded in source
1
Scheduled or as-needed release trigger
trigger
“Every quarter or as needed, we can create a new version and ship it immediately.”
2
MLflow tracks artifacts and versions
integration
“This helped the data science team reduce the amount of time they spent tracking artifacts, model versions and notebooks”
3
Model Serving deploys and monitors models
integration
“a comprehensive service used for deploying, managing and monitoring machine learning models, whether they are developed by Databricks or sourced from other providers.”
4
ETL, NLP, and classification tasks
ai_action
“their data teams used these ML models to support standard extract, transform and load (ETL) pipelines to ingest data from various sources, perform NLP tasks (e.g., summarizing and sentiment analysis) and execute binary classification wor…”
5
Increased annual model output
output
“FiscalNote can now deploy 3x the number of models annually, compared to their output before Databricks.”
Reported outcome

Using MLflow and Databricks Model Serving, FiscalNote made model deployment 3x faster, increased analyst productivity, and now deploys 3x the number of models annually compared to before Databricks.

Reported metrics
Model deployment speed3x faster
ML models deployed annually3x
deployment cadence before Databricksonce a year
Analyst productivityincreasing analyst productivity
Reported stack
DatabricksMLflowDatabricks Model Serving
Source
https://www.databricks.com/customers/fiscalnote
Read source ↗

Frequently asked questions

What did this team achieve with this AI workflow?

Using MLflow and Databricks Model Serving, FiscalNote made model deployment 3x faster, increased analyst productivity, and now deploys 3x the number of models annually compared to before Databricks.

What tools did this team use?

Databricks, MLflow, Databricks Model Serving.

What results were reported?

Model deployment speed: 3x faster; ML models deployed annually: 3x; deployment cadence before Databricks: once a year; Analyst productivity: increasing analyst productivity (source-reported, not independently verified).

How is this back office ops AI workflow structured?

Scheduled or as-needed release trigger → MLflow tracks artifacts and versions → Model Serving deploys and monitors models → ETL, NLP, and classification tasks → Increased annual model output.

Related back office ops cases