How FactSet cut model training and deployment from days to hours with Databricks AI Runtime
FactSet accelerated its Text-to-Formula (TTF) pipeline by moving model training and deployment onto Databricks AI Runtime, shrinking the end‑to‑end cycle from multiple days to just a few hours. The new runtime leverages Databricks‑optimized Spark clusters and integrated model serving, enabling rapid iteration on large‑scale financial text data. Engineers can now retrain and push updated formula‑generation models on a shared platform without bespoke infrastructure. This shift dramatically reduces time‑to‑value for financial analytics applications.
⚡ Key Takeaways
- Training and deployment time reduced from days to hours after adopting Databricks AI Runtime.
- Uses Databricks‑optimized Spark clusters for distributed model training.
- Achieves faster iteration cycles, cutting operational latency for model updates.
- Integration is done via Databricks AI Runtime’s built‑in model serving APIs.
- Requires migration of data pipelines to Databricks environment; on‑prem setups are not compatible.
🔧 Tools & Libraries
Cutting model turnaround from days to hours lets financial data teams iterate quickly, keeping formula generation models up‑to‑date with market changes and delivering fresher insights to end users.
✅ Practical Steps
- Migrate your training data and code to a Databricks workspace and configure a Spark cluster with the Databricks AI Runtime.
- Use the Databricks AI Runtime model serving API to deploy the trained TTF model directly from the training notebook.
- Monitor training and serving metrics via Databricks’ built‑in dashboards to validate performance gains.
Want the full story? Read the original article.
Read on Databricks Blog ↗