Patents Examined by Eddy Cheung
-
Patent number: 12783518Abstract: Embodiments herein describe systems and techniques enabling effective use of tokenization for analysis and processing of text in natural language processing tasks. A disclosed system includes a text classifier model to perform predictions based on tokenized data and training data. In an embodiment, the text classifier model is used to generate a score for sentence tokens to be identified as a positive match to a query. In response to the text classifier model reaching a threshold score of prediction for a sentence, the sentence is tagged as a target sentence. A pre-defined window search of neighbor tokens is performed to determine a region of iterative similarity scores computed based on text from the knowledge base and a span of text from a document that includes the sentence, and an effective text span for a predicted data point is obtained based on the region of iterative similarity scores.Type: GrantFiled: June 27, 2025Date of Patent: November 3, 2026Assignee: INTERNATIONAL BUSINESS MACHINES CORPORATIONInventors: Valdir Salustino Guimaraes, Rodrigo Reis Alves, Angelo Moore, Vasanthi M. Gopal
-
Patent number: 12783787Abstract: Techniques for logical text passage generation and retrieval for retrieval-augmented generation. The techniques involve processing markup language documents to generate logical text passages and their corresponding embeddings. These embeddings are indexed for efficient retrieval. Upon receiving a user utterance, a user query is formed and transformed into an embedding to query the index. Relevant text passages are identified and used to prompt a large language model (LLM), which generates a completion. This completion is then sent as a response to the user. The process effectively bridges user queries with relevant information through advanced embedding and natural language processing techniques, enabling accurate and contextually appropriate interactions within a user-agent dialogue framework.Type: GrantFiled: March 25, 2024Date of Patent: November 3, 2026Assignee: Amazon Technologies, Inc.Inventors: Shahebaj Mahemood Pathan, Ankit Ashvin Bhodia, Vinayshekhar Bannihatti Kumar, Sopan Khosla, Ruiyang Li, Ramya Kota Ural, Paramvir Singh, Rashmi Gangadharaiah, Kavisha Arora
-
Patent number: 12775671Abstract: Various implementations are directed towards fine-tuning a large language model (LLM) using search engine feedback (e.g., responsive content generated based on a reference source material such as a set of search engine results). Additionally or alternatively, a supervision signal can be generated based on comparing search engine conditioned LLM output with unconditioned LLM output. In many implementations, the supervision signal(s) can be used in training a reward model using reinforcement learning, where the trained reward model can be used in fine-tuning the LLM.Type: GrantFiled: September 11, 2025Date of Patent: October 27, 2026Assignee: GOOGLE LLCInventors: Hyun Jin Park, Changwan Ryu
-
Patent number: 12769787Abstract: Autonomous metadata harmonization in multi-cloud environments (e.g., using a computerized tool), is enabled. For example, a corresponding process can comprise validating, using a local artificial intelligence model that was trained using federated learning applied to a group of zones of a multi-cloud architecture, a metadata attribute in a zone of the group of zones of the multi-cloud architecture, in response to determining a deviation in the metadata attribute, determining a trust score, applicable to the metadata attribute, using a defined probabilistic model, in response to the trust score being determined to satisfy a defined trust threshold, queuing a metadata update applicable to the metadata attribute, using a defined weighted priority process, determining a priority order for the metadata update among a group of metadata updates applicable to the group of zones of the multi-cloud architecture, and based on the priority order, initiating the metadata update.Type: GrantFiled: September 19, 2025Date of Patent: October 20, 2026Assignee: DELL PRODUCTS L.P.Inventors: Avinash Kumar, Zijia Wang, Srinath Kappgal
-
Patent number: 12754808Abstract: Techniques are provided for implementing asynchronous region-aware caching for snapshot prefetching and restore. A first application within a first container of a container orchestration platform may be hosted at a first region of a multi-cloud environment. A second application within a second container of the container orchestration platform may be hosted at a second region of the multi-cloud environment. Data of the first application may be stored within storage located within the first region of the multi-cloud environment. In response to determining that the second application has a threshold probability of accessing the data of the first application (e.g., a snapshot that the second application can restore), the data may be cached as cached data within a network edge of the second region for access by the second application.Type: GrantFiled: November 25, 2024Date of Patent: October 6, 2026Assignee: NetApp, Inc.Inventor: Madhuvanthi Janarthanam
-
Patent number: 12755044Abstract: A method for enabling continuity of access to a primary dataset stored in a computer network is provided. The method utilizes a node, a computer processor, and non-transitory computer-readable media storing computer-executable instructions. The node is connected to the computer network. The method includes configuring the mobile device for wireless connection to the computer network. The method includes configuring the mobile device to receive indication that the node is disconnected from the computer network and transmit a copy of the primary dataset or a portion thereof from the computer network to the node. The method includes configuring the node to enable operations on the dataset copy and keep a record of them. The method includes configuring the node to receive indication that the node is connected to the computer network and transmit the record to the computer network.Type: GrantFiled: June 27, 2025Date of Patent: October 6, 2026Assignee: Bank of America CorporationInventors: Ion Freeman, Manu Kurian, Ganesh Bonda, Aeric Solow, Sasikumar Purushothaman
-
Patent number: 12748751Abstract: Techniques are disclosed for managing metadata of a distributed database system in a hybrid manner. A computer system may receive, from a computing device, a request to access a set of data stored in nodes of a distributed storage system that is a caching layer of the system. The system retrieves metadata for a set of data specified in the request, including accessing a reversemap storing a reverse-ordered copy of original metadata stored in a metadata store of the system, where the reversemap is stored on durable storage of the system. Based on retrieving the metadata from the reversemap, the system accesses nodes of the distributed storage system, where the reversemap specifies the nodes of the distributed storage system that store the set of data. The system transmits, to the computing device, information indicating a result of accessing data stored in nodes of the distributed storage system.Type: GrantFiled: June 25, 2025Date of Patent: September 29, 2026Assignee: Salesforce, Inc.Inventors: Venkateswararao Jujjuri, Sushanth Rai, Jayant Kumar, Anup Ghatage
-
Patent number: 12748744Abstract: A method of configuring a Log-Structured Merge (LSM)-tree structure for a key-value database is provided. The LSM-tree structure having a series of levels. The method includes: obtaining a value for a level capacity parameter associated with a largest level of the series of levels of the LSM-tree structure; determining values for level capacity parameters associated with intermediate levels, respectively, of the series of levels based on an optimal cost tradeoff between a range lookup operation cost and an update operation cost associated with the LSM-tree structure based on the obtained value for the level capacity parameter associated with the largest level; and configuring the LSM-tree structure based on the determined values for the level capacity parameters associated with the intermediate levels. There is also provided a corresponding system for configuring an LSM-tree structure for a key-value database and a corresponding LSM-tree-based key-value database system.Type: GrantFiled: May 2, 2025Date of Patent: September 29, 2026Assignee: Nanyang Technological UniversityInventors: Siqiang Luo, Junfeng Liu, Fan Wang
-
Patent number: 12743349Abstract: Various implementations implement a standby directory service. For example, an example process may include obtaining user data associated with an activity of a user via a sensor in a physical environment. The process may further include, at a device including one or more processors, determining that an interruption event has occurred at a host directory service, wherein the device and the host directory service are communicatively coupled via a first network protocol associated with a first network, and in response to determining that the interruption event has occurred at the host directory service, implementing a standby directory service, wherein the standby directory service is generated and updated via a second network protocol that is different than the first network protocol.Type: GrantFiled: July 18, 2025Date of Patent: September 22, 2026Assignee: CAYOSOFT, INC.Inventors: Andrei Polevoi, Robert John Bobel, III, Alexander Vadimovich Tsvetkov
-
Patent number: 12743347Abstract: Techniques for data recovery are provided. The techniques include receiving a notification message sent by a computing node in a distributed graph computing system, where the notification message represents that the corresponding computing node successfully stores current checkpoint data in a persistent memory; generating a checkpoint identifier corresponding to a current checkpoint when determining that all computing nodes in the distributed graph computing system successfully store the current checkpoint data in respective persistent memories; and sending the checkpoint identifier to each computing node, so that each computing node associates the checkpoint identifier with the current checkpoint data stored in the persistent memory, and the associated current checkpoint data is used for data recovery during recovery execution before the corresponding computing node obtains a checkpoint identifier corresponding to a next checkpoint.Type: GrantFiled: July 22, 2024Date of Patent: September 22, 2026Assignee: Beijing Volcano Engine Technology Co., Ltd.Inventors: Miaomiao Cheng, Jiujian Chen, Cheng Zhao, Cheng Chen, Yongmin Hu, Xiaoliang Cong, Liang Qin, Hexiang Lin, Ronghua Li, Guoren Wang, Shuai Zhang, Lei Zhang
-
Patent number: 12730804Abstract: The present disclosure relates to systems, non-transitory computer-readable media, and methods that utilize a meta-prompting model to generate and store modified system prompts for agent models. For example, the system utilizes an agent model to produce a generated task output from a system prompt corresponding to a task input. The system utilizes a meta-prompting model to compare the generated task output with an expected task output and determine one or more task output differences. The system utilizes the meta-prompting model to generate a modified system prompt for the agent model to produce a task output aligning with the expected task output according to the task output differences. In some embodiments, the system iteratively adjusts the modified system prompt. In some embodiments, the system stores the modified system prompt in a system prompt repository for later use with the agent model for a particular task.Type: GrantFiled: December 23, 2025Date of Patent: September 8, 2026Assignee: Dropbox, Inc.Inventors: Eric Wang, Dmitriy Meyerzon, Hans Sayyadi
-
Patent number: 12724676Abstract: A storage manager for an information management system determines whether one or more predetermined conditions have been met for transferring metadata of previously performed backup jobs stored in a first management database. A backup job may correspond to a backup operation of a primary storage device of a first client computing device. In response to a determination that one or more of the predetermined conditions have been met, the storage manager may transfer metadata for a second plurality of backup jobs to a second management database of a recovery manager. The recovery manager may receive a request to restore data to the primary storage device of the first client computing device based on the metadata of the second plurality of backup jobs. A media agent managed by the recovery manager may then restore the requested data to the primary storage device of the first client computing device.Type: GrantFiled: July 1, 2025Date of Patent: September 1, 2026Assignee: Commvault Systems, Inc.Inventor: Prasad Nara
-
Patent number: 12724799Abstract: System, method, and various embodiments for a tabular data integrity and query processing system are described herein. An embodiment operates by receiving a query to be executed against a knowledgebase. One or more keywords are identified from the query, and a vector search is performed against the knowledgebase based on the one or more keywords, the knowledgebase including documents that have been divided into a plurality of chunks. A subset of chunks related to generating an answer for the query are identified based on the vector search, the subset including a first chunk with a table ID. A table image corresponding to the table ID is identified. A prompt is generated instructing a language model to generate the answer to the query based on the subset of chunks, including the first chunk and the table image. The answer is provided.Type: GrantFiled: May 23, 2025Date of Patent: September 1, 2026Assignee: SAP SEInventors: Kuluhan Binici, Wei Liang Lim, Hu Soon Tan, Zhen Shu, Gopal Jhunjhunwala
-
Patent number: 12717777Abstract: Systems and methods for natural language query processing and visualization. In embodiments, a structure associated with a dataset and a natural language question are obtained and provided to an AI model to request, from the AI model, a query that may be used for retrieving data from the dataset responsive to the natural language question. The query is received form the AI model and executed against the data in the dataset to retrieve data responsive to the natural language question. In embodiments, the data responsive to the natural language question is analyzed to determine one or more structural characteristics of the data responsive to the natural language question, and a graphical visualization of the data responsive to the natural language question is generated based on the one or more structural characteristics of the data responsive to the natural language question.Type: GrantFiled: June 17, 2025Date of Patent: August 25, 2026Inventors: Aaron C. Cunningham, Ronald Raymond Austring, Louis A. Martinez, John B. Wheeler
-
Patent number: 12711100Abstract: Techniques are described for incrementally determining checksums for a snapshot. An example method comprises identifying, by a data platform implemented by a computing system, a plurality of leaf nodes and a plurality of intermediate nodes in tree data corresponding to a snapshot of a storage system at a particular time, wherein the intermediate nodes each comprise one or more pointers identifying one or more of the leaf nodes, and the leaf nodes each include an indication of file system data of the storage system. The method includes determining, by the data platform, a checksum for each of the leaf nodes, determining, by the data platform, a checksum for each intermediate node based on the checksum of the one or more leaf nodes identified by the pointers of the intermediate node; and storing, by the data platform, the checksum for each of the leaf nodes and each of the intermediate nodes.Type: GrantFiled: May 9, 2025Date of Patent: August 18, 2026Assignee: Cohesity, Inc.Inventors: Anirudh Kumar, Namit Sharma, Avinash Nath Aita, Samanvitha Reddy Panyam
-
Patent number: 12705144Abstract: A method may include, based on a recovery policy associated with the primary database, creating a first snapshot of the primary database at a primary computing system, replicating the first snapshot of the primary database to a secondary computing system, wherein a backup database of the primary database is hosted on the secondary cluster, in response to the primary database being unavailable, applying the first snapshot of the primary database to the backup database, creating, according to the recovery policy associated with the primary database, a snapshot of the backup database, in response to the primary database having restored availability, creating, according to the recovery policy associated with the primary database, a second snapshot of the primary database.Type: GrantFiled: May 28, 2025Date of Patent: August 11, 2026Assignee: Nutanix, Inc.Inventors: Adil Gul Rather, Akshay Ashok Chandak, Altaf Mahmood, Jagadeesh Kumar Dittakavi
-
Patent number: 12699634Abstract: Methods, systems, and devices for data management are described. A data management system (DMS) may support backup and recovery of a filesystem. The DMS may, as part of the backup, obtain a snapshot of the filesystem. The DMS may mount the snapshot within a user space of a node of the DMS. The user space may be isolated from other processes that execute on the node of the DMS. The DMS may generate, within the user space based on the mounted snapshot, index files that index the filesystem. In response to a subsequent request to access one or more files, the DMS may retrieve the one or more files from the snapshot mounted in the user space based on the index files.Type: GrantFiled: June 23, 2025Date of Patent: August 4, 2026Assignee: Rubrik, Inc.Inventors: Anshul Mohan Gupta, Takeshi Suzuki, Lewi Abseno
-
Patent number: 12699740Abstract: A generative document system creates generative search results documents using generative artificial intelligence (AI) models and dynamically determines which one of the generative search results documents to provide in response to a search query. For example, in response to receiving a search query, the generative document system obtains search link results (e.g., website links and corresponding grounding information) for the search query and utilizes this information with multiple generative AI models to generate various types of generative search results documents. Additionally, the generative document system generates and utilizes a generative document arbitration model to determine, based on the search link results, which of the generative search results documents to provide in response to the search query.Type: GrantFiled: March 28, 2025Date of Patent: August 4, 2026Assignee: Microsoft Technology Licensing, LLCInventors: Mohamed Salman Ismail Gadit, Samer Hassan Hassan, Andrew Peter Oakley, Nathan James Chalmers, Ronak Ashwinkumar Shah, Doran Chakraborty, Aparna Rajaraman, Lile Palma Hattori, Rami Sayar, Vera Aster Brubaker, Matthew Yoshimi Cedeno
-
Patent number: 12693943Abstract: Techniques are provided for performing a storage operation targeting objects stored across multiple storage tiers of a cloud storage environment. A volume may be backed up as objects stored across the multiple storage tiers of the cloud storage environment, such as a standard storage tier directly accessible to the storage operation, an archival storage tier not directly accessible to the storage operation, etc. The storage operation may target the objects, such as where the storage operation is a directory restore operation to restore a directory of the volume. The storage operation can be successfully implemented such as to restore the directory even though objects of the storage operation are stored across the multiple storage tiers of the cloud storage environment.Type: GrantFiled: May 12, 2025Date of Patent: July 28, 2026Assignee: NetApp, Inc.Inventors: Sanhita Praveen Dhavale, Tijin George, Avanthi Rajan, Harsha Vardhan Reddy Perannagari, Sumedh Guha
-
Patent number: 12694027Abstract: Embodiments of the disclosed technologies include parsing a query into a first query portion and at least one second query portion, matching an embedding of the at least one second query portion with an embedding that corresponds to a portion of a document of a document set, mapping the portion of the document to a first node of a graph; by a generative artificial intelligence model, constructing a graph query based on at least the first node, executing the graph query on the graph to identify a second node of the graph, extracting a path from the graph, and configuring the path for output at a device.Type: GrantFiled: May 6, 2025Date of Patent: July 28, 2026Assignee: Microsoft Technology Licensing, LLCInventors: Zhentao Xu, Mark Jerome U. Cruz, Matthew R. Guevara, Tie Wang, Manasi R. Deshpande, Xiaofeng Wang, Zheng Li