Understanding the technology behind Sisu
Our platform combines automatic feature engineering, model selection, and a novel query engine to rapidly find useful insights in cloud-scale data.
Proven at scale with industry leaders
Technology tested and deployed to answer key business questions from the world’s largest datasets.
Purpose-built to find insights, fast
- Declarative query model
- Efficient search and featurization
- Automated model and feature selection
- Personalized ranking and relevance
- New interface to data
Declarative query model
Like a search engine that surfaces relevant results from billions of possibilities, Sisu starts with your critical business metrics and automatically explores all your data to identify the highest-impact populations.
Efficient search and featurization
Sisu applies novel search algorithms to prioritize analysis of the most promising columns in the data. Our platform rapidly scans the schema, data types, cardinality, and related features, and then performs automated transformations to expose factors of statistical interest by:
- Grouping continuous attributes
- Computing higher-order and conjunctive features
- Keyword analysis including n-gram extraction
Automated model and feature selection
Sisu performs a model and feature search to identify statistically significant factors for both point-in-time and time-based analyses. This requires modifications for scalability and usability, including:
- Hyperparameter tuning to balance feature complexity and accuracy
- Prioritized search with automated FDR controls
- Data-aware parallel and columnar execution
Personalized ranking and relevance
Our platform utilizes past behavior as well as several other statistical signals to continuously improve its ranking and relevance of results.
New interface to data
To accelerate decision-making, Sisu generates the top insights with clear explanations of population behavior and the impact on each metric.
Designed for cloud-scale data
Sisu is the only analytics platform built for the speed and scale of cloud-native warehouses. Designed for highly-parallel processing, the platform quickly uncovers valuable interactions between subpopulations and metrics that humans don't have the time or processing power to find.
