HubSpot's feature store architecture is organized into a discovery catalog, an online store, and an offline data lake. Apache HBase delivers low-latency lookups for real-time model inferences, while Apache Iceberg maintains the offline repository for batch training. A Spark streaming pipeline captures HBase updates and stages them on S3, allowing a daily loader to generate wide point-in-time rows in Iceberg that eliminate complex joins during training.
Key points
Teams can access existing SQL tables, microservices, and CRM properties directly through…: Teams can access existing SQL tables, microservices, and CRM properties directly through a single unified interface without duplicating raw data.
Automated synchronization captures HBase write events to ensure consistent feature values…: Automated synchronization captures HBase write events to ensure consistent feature values across both training models and production inference.
Daily loader jobs generate wide point-in-time entity rows within Iceberg to bypass resour…: Daily loader jobs generate wide point-in-time entity rows within Iceberg to bypass resource-heavy joins during model training runs.
Sources checked by our analysis pipeline: 1
Written by Leena Bhandari, Senior Software Engineer II, Andrea Cicalese, Senior Software Engineer II, and Durai Ramasubramanian, Senior Software Engineer II at HubSpot.
Why is fragmented data a bottleneck for engineering teams?
Every machine learning model runs on features which are signals it learns from and reasons over. In a large organization they are scattered across operational databases, warehouse tables, event streams, and the pipelines that individual teams have built. An ML engineer starting a new model first has to find the signals they need, make sense of the data, then write one-off code to extract and transform it. The next team that needs an overlapping signal builds it all over again. This makes shipping a new functionality slow and hard. The need for a unified store has sharpened as teams build LLM and agent powered features needing a shared, trustworthy place to find and serve context at scale.
A Feature store is a central layer where features are defined once, then discovered and reused across teams and across the training-to-serving boundary. But in a large organization, most of that…
The neocloud infrastructure provider is accelerating its aggressive capital spending to expand global capacity amid growing demand for artificial intelligence hardware. Robust long-term commitments are helping counter recent industry turbulence driven by macroeconomic pressures and an investigation surrounding key partner Nvidia. However, the firm faces mounting financial exposure due to high borrowing costs and substantial existing debt obligations.
The artificial intelligence company shared its latest financial metrics directly with backers during a Tuesday update. This notable surge reflects a rapid turnaround from the previous quarter's more modest pace, demonstrating escalating commercial demand. The substantial figures highlight the organization's expanding market footprint as it scales its operations.
Surging cloud infrastructure consumption from major technology clients helped the Nvidia-backed firm counter recent market skepticism over aggressive capital spending. Despite carrying substantial debt and facing broader industry delays, the company raised its annual forecast as sustained demand outpaced Wall Street expectations. The positive results sparked an immediate share rebound following weeks of sharp market declines.