Data provides the means to answer critical questions in government, however, the ability to gain knowledge is plagued with problems and setbacks resulting in large datasets being unused. Agencies with large amounts of unstructured data struggle with the time required to analyze, organize, and operationalize information.
To address these problems, InterImage developed a web-based, cloud native machine learning (ML) tool, MLCircuit, to accelerate the ML model development process. MLCircuit enables rapid development of ML-based models, saving 70% of initial model development time, via a low/no-code GUI. Models are developed iteratively and tested using automated features of MLCircuit such as autogenerated APIs. MLCircuit supports a wide variety of ML tasks such as topic modeling, sentiment analysis, classification, regression and anomaly detection. MLCircuit automates the common patterns in every step of the ML process including data ingestion, data maintenance, model training, model evaluation and model deployment. Key features of MLCircuit include:
- Standardized data ingest and analysis tasks including ETL
- Natively integration with common data sources such as APIs and SQL databases while allowing users to upload their own custom datasets
- Abstractions around storage mechanisms to determine the best way to store and maintain the dataset
- Standard and modular implementations of common ML algorithms
- Drag and drop to chain various nodes together to form dynamic and complex flow of data between the various components
- Easily swap out different portions of a ML pipeline to test new solutions – a process that would otherwise be time consuming and resource intensive using traditional methods
- Orchestrates training, versioning, evaluation and visualization of models so the user can focus on making meaningful optimizations to benefit performance.
- Deployment engine automatically generates an API around a specific version of a model
- Handles common issues such as data validation, scalability and security
- Easily test novel data against deployed models