Solution
SūmerSports rebuilt its football-analytics stack on the Databricks Data Intelligence Platform, consolidating fragmented Azure-native services, open-source tools, and custom orchestration into a single foundation that handles ~200TB of structured and semi-structured data per NFL season from 14 sources (sensor tracking, scraped web data, in-house scouting evaluations, client data, NFL contract financials). Unity Catalog governs the data sources with curated, tagged datasets and lineage; MLflow tracks experiments and prototypes; Databricks Asset Bundles let data scientists deploy model training, cleaning, and inference jobs as infrastructure-as-code; ~95% of workloads run on serverless compute. AI/BI Dashboards, Genie, and the Databricks Assistant surface analytics to ~60 internal users and to customers. Embeddings for the SūmerBrain chatbot are prototyped on Mosaic AI and served in production through AWS Bedrock. Time-to-market dropped from 3 months to 4 weeks, and post-game insights from 4 days to under a minute.
Data flow
SūmerSports ingests ~200TB/season of structured and semi-structured football data from 14 sources (sensor telemetry, scraped web data, in-house scouting evaluations, client data, NFL contract financials) staged in AWS S3, runs nightly jobs to clean, feature-engineer, retrain, and serve inference, and stores everything under Unity Catalog as curated, tagged datasets. ML development happens in collaborative Notebooks tracked by MLflow; Asset Bundles deploy training/cleaning/inference jobs as infrastructure-as-code. AI/BI Dashboards and Genie surface analytics internally and to customers. Embeddings prototyped in Mosaic AI power the SūmerBrain chatbot, served in production through AWS Bedrock. ~95% of jobs run on serverless compute.
Solution architecture
8 components · 4 layers - Compute
- Databricks Data Intelligence Platform Single platform underpinning ~200TB/season of football-data ingestion, feature engineering, model training, and inference across ~14 data sources
- Agent Bricks / Mosaic AI Prototypes the embeddings powering SūmerBrain chatbot interfaces (production models served via AWS Bedrock)
- Databricks Assistant Context-aware AI assistant integrated with Unity Catalog so analysts can ask routine table/availability questions without involving data engineering
- Databricks Serverless Compute ~95% of workflows run on serverless, eliminating cluster management for most jobs
- Serving
- Databricks AI/BI Collaborative Notebooks and Dashboards surfacing analytics internally and to customers, with Genie enabling natural-language access
- Orchestration
- Databricks Asset Bundles Infrastructure-as-code templates that let data scientists deploy model training, cleaning, and inference jobs to production themselves
- Governance
- Unity Catalog Governs 14 primary data sources with curated, tagged datasets, lineage, and access control across product teams
- MLflow Experiment tracking, exploratory analysis, and lifecycle management for ML projects in a single persistent location
- Asset Bundles as infrastructure-as-code for model deployment
- MLflow for experiment tracking and lifecycle management
- Mosaic AI for embeddings and chatbot prototyping
- Production model serving via AWS Bedrock
- Runs on AWS
- Single Databricks Data Intelligence Platform across ingestion, ML, and analytics
- Unity Catalog over 14 curated data sources
About 95% of our workflow is on serverless. It’s one of the biggest boons in our day-to-day operations.
In the last six months, we’ve cut it down to where it’s happening within the minute
We were able to serve customers that EPA model in just under four weeks. The same work would have previously taken three months.
