Our Proprietary Automation Technology
As part of Astreya’s Service Desk Solution, we utilize Astreya technologies to receive, triage, and manage user support requests submitted via EMail and Chat. This allows Service Desk to automate and deliver scalable IT self-help (Tier 0) and self-service (Tier 1).
Our platform unifies the broad array of AI and ML capabilities to provide unique data-driven IT service delivery and support including:
- Document retrieval for search engine
- Question answering for search engine text highlights
- Natural language understanding for chatbot clarifying questions
- Decision trees for chatbot clarifying questions
- Finite state machines for automations
We Accelerate IT User Support Automation
Our proprietary analytics platform represents a unique ability to manage, curate, operationalize and significantly simplify the workflow needed to accelerate AI time-to-value.
- Sourcing: Native JDBC connector allows for connectivity to virtually any data source that has a JDBC driver.
- Transformation: With a built in SQL editor, write and execute queries directly against data sources for analysis, process mining and anomaly detection.
- Augmentation: Append query results with inferences from your own or third party ML models.
- Visualization: Aggregate, group and filter augmented query results with visualizations.
Map fields in a transaction or log data to a case id, process step, and timestamp. A process diagram will automatically be rendered from the data allowing the analyst to observe and discover their business processes:
- Performance: Key statistics are calculated for the entire process, as well as nodes and edges in the process diagram.
- Statistics: Frequencies, durations, relative frequencies, rework, and active cases. Monitor when these statistics exceed a defined threshold.
- Deviation: Define an ideal path for the process based on desired business practices. Observe the most common path through a process based on actual activity. Monitor and notify when cases deviate from the ideal or most common paths.
- Variation: Compare statistics and process flows across cohorts of cases to perform A/B testing. Monitor cohorts and notify stakeholders when variation occurs.
Performed on raw, summarized, and process-oriented data. The system monitors datasets and alerts to anomalies using built-in models, custom algorithms, or third-party services. The built-in anomaly detection models provided by Astreya’s analytics platform are:
- Simple Threshold: Metrics are compared to a static threshold (above, below, outside & between).
- Statistical: Metric values are compared to some number of standard deviations from the mean of the sample from which they were drawn.
- Time Series: An ARIMA (Autoregressive Integrated Moving Average) model is trained on time series data. The most recent data point (held out) is compared to its respective prediction from the trained model.
- Feature Splitting: An Isolation Forest model is trained on a dataset to determine the split dataset’s tree structure. New records are compared to the tree structure to determine if they are anomalies.
- Dimensionality Reduction: A PCA (Principal Component Analysis) algorithm is trained to decompose a dataset to its core components. New records are compared to the decomposed dataset. Reconstruction error is measured to identify anomalies. Once models are trained, they can be scheduled to run against new data streams to infer anomalies automatically.