US 12143425

US patent 12143425, titled "Rapid predictive analysis of very large data sets using the distributed computational graph," was invented by Jason Crabtree and Andrew Sellers. The application was filed on July 21, 2024, and the patent was granted/issued on November 12, 2024. The current assignee is Qomplx Inc.

Abstract:
The patent describes a system for the predictive analysis of very large datasets using a distributed computational graph. It includes a data receipt software module for receiving streaming data from various sources. In a batch data pathway, a data formalization module formats the input data for storage. A batch event analysis server then inspects this stored data for trends, patterns, or knowledge, and aggregates the data for a message handler. In parallel, a streaming pathway utilizes a transformation pipeline software module to manipulate the data stream. A system sanity software module monitors system performance, receives status updates from the message handler, optimizes the system, and retrains the transformation pipeline based on insights from both streaming and batch analysis.

Plain-language overview of independent claims:

The patent describes both a system and a method for rapid predictive analysis of large datasets using a distributed computational graph.

Independent System Claim:
The system is designed for rapid predictive analysis of very large data sets. It comprises several software modules working together:

1. Data Receipt Module: Gathers incoming data streams from various sources.
2. Data Filter Module: Cleans the incoming data by removing incomplete, damaged, or incongruent records, then splits the clean data into two identical streams. One stream goes to data formalization, and the other to the transformation pipeline.
3. Data Formalization Module: Formats the data stream for consistent storage.
4. Input Event Data Store: Stores the formatted data for long-term access, retrieval, and analysis.
5. Batch Event Analysis Server: Analyzes the stored data for trends, historical events, or cause-and-effect relationships, summarizing these findings and interacting with the messaging module.
6. Transformation Pipeline Module: Processes the real-time data stream received from the filter, applying various functions and providing results back to the system, while also receiving directives for modification from the system sanity and retrain module.
7. Messaging Software Module: Acts as a central communication point, receiving administrative directives, summaries from batch analysis, and results from the transformation pipeline, and sends status and execution directives to the system sanity and retrain module.
8. System Sanity and Retrain Module: Monitors the system's stability and progress, comparing information against set parameters. It adjusts the operational behavior of other modules to maintain required function, issues alerts for degraded status, and implements administrative changes.
9. Output Module: Prepares and routes analyzed information to external destinations for further action.

Independent Method Claim:
The method for predictive analysis of large data sets involves the following steps:

1. Receive Streaming Input: Obtain continuous data from multiple sources.
2. Filter Data: Clean the received data by removing anything incomplete, misconfigured, or damaged.
3. Formalize Input Data: Standardize the format of the data for both batch processing (historical analysis) and streaming analysis.
4. Perform Data Transformations: Apply one or more functions to the formalized data.
5. Perform Sanity Checks and Retraining: Validate the results from the real-time transformation pipeline and adjust the analysis process based on findings from the batch analysis of historical data.
6. Output Results: Present the final analysis results in a predefined format.

As of April 26, 2026, no specific dockets for patent number 12143425 were found in the CAFC 2026 dockets.