Other

Vijay Madisetti

Madisetti, Vijay

Bahga, Arshdeep

Arshdeep Bahga

Jae Yoon Kim

Avinash Rao

Michael W Richter

Systems and methods for securing agent communications in a multi-agent system including deploying an agent to communicate with external servers, instantiating a sidecar security service in communication with the agent and including at least one of a guardrails service, a security layer, an encryption module, and an integrity checker configured to validate resources tools using hash-based verification mechanisms. Communications including messages between the primary and the external servers are intercepted by the sidecar security service, which then performs at least one of filtering the one or more messages by the guardrails service, authenticating one or more server connections by the security layer, encrypting one or more outbound requests by the encryption module, or verifying an integrity of resources and tool definitions by the integrity checker.

Sidecar security pattern for agent communications

A method for utilizing hierarchical tokens (h-tokens) in large language models (LLMs) including receiving and processing an input prompt to generate tokens, identifying functional components derived from a subset of tokens, associating or generating h-tokens for at least one of the functional components by compressing the subset of tokens into an h-token for each functional component, associating or generating implementation characteristics including a defined function, an identified event, and/or an implementation method for each h-token, performing a retrieval augmented generation (RAG) processing function on each h-token, and generating a response from an output of the RAG processing function.

Method and system for multi-level artificial intelligence supercomputer design

A method of performing secure write operations including receiving a secure write instruction at a programmable-logic controller (PLC), performing a validation operation on the secure write instruction at the PLC, rejecting the secure write instruction responsive to determining the secure write instruction is invalid, and responsive to determining the secure write instruction is valid, verifying a permissibility of the secure write instruction at the PLC by comparing one or more parameters of the secure write instruction to a range of permitted values stored in the PLC, rejecting the secure write instruction responsive to determining the secure write instruction is impermissible, and executing the secure write instruction at the PLC responsive to determining the secure write instruction is permissible.

Method and system for securing SCADA and industrial control software through secure-write operations

A method for agent failure detection including listening for an agent heartbeat signal from an agent, updating a most recent heartbeat timestamp for the agent upon receiving the agent heartbeat signal from the agent or, upon not receiving the agent heartbeat signal, probing the agent for a response. Upon receiving the response from the agent, the suspect agent designation is removed from the agent and the most recent heartbeat timestamp is updated. Upon not receiving the response, the method further includes designating the agent as a failed agent, sending a message to a service mesh and/or a message pool regarding the agent being designated a failed agent, activating a shadow agent associated with the agent, and resuming operation with the shadow agent replacing the agent.

Fault tolerant multi-agent generative AI applications

A system for processing LLM operations including a memory module, a data receiver to receive an input data stream and break the input data stream into batches of input data, a processing layer to process the batches of input data in the one or more memory modules and having LLMs to process the batches of input data and generate corresponding batches of processed output data and an output broker to cache results from the plurality of LLMs in a results cache, assign priority scores to the cached results, rank the cached results based on their priority scores, and provide ranked results responsive to subsequent similar input data received by the data receiver from the results cache.

Systems and methods for countering artificial intelligence-driven disinformation attack, including collecting data from data sources, preprocessing the data, identifying a disinformation pattern in the preprocessed multimodal data, performing a network analysis and a historical disinformation comparison by comparing the disinformation pattern to a historical disinformation campaign database comprising historical disinformation patterns on the disinformation pattern, and determining a threat level tier for the disinformation pattern, performing a first threat response comprising responsive to determining a first threat level tier, performing a second threat response responsive to determining a second threat level tier, and performing a third threat response responsive to determining a third threat level tier.

Method and system for countering AI-driven disinformation with defensive actors

A method for generating adversarial data for use in an LLM including receiving an input condition including authentic data records, extracting by a PII extraction module an authentic data record from the input condition, assigning a utility score to each data field of the plurality of data fields of the authentic data record by a generator neural network using a utility function, either receiving an adversarial loss term indicating a target divergence between the authentic data record and a synthetic data record and generating the adversarial loss term, and generating the synthetic data record comprising a plurality of synthetic data fields based on the input condition, the utility score of each data field, and the adversarial loss term, each synthetic data field of the plurality of synthetic data fields being associated with a data field of the plurality of data fields.

Method and system for protecting and removing private information used in large language models

A method of configuring local secure memory for a programmable logic controller (PLC) including verifying a content of the local secure memory by comparing the content of the local secure memory to a verification profile, initiating a secure recovery process responsive to unsuccessfully verifying the content of the local secure memory, and, responsive to successfully verifying the content of the local secure memory, adding at least one of an approved operation or a security parameter to the content of the local secure memory and updating the verification profile responsive to adding the at least one of an approved operation or a security profile.

A system and method for generating risk assessment scores for blockchain addresses and smart contracts, including constructing a graph representation having nodes representing one of a blockchain address or a smart contract address on a blockchain and edges representing a transaction between an externally-owned blockchain account on the blockchain network and a smart contract address on the blockchain network, applying pattern recognition algorithms to the graph representation to identify sub-graphs having nodes indicating relationship structures indicative of potential fraudulent activity, executing filtering that analyze transaction flows, logic patterns, and behavioral characteristics of nodes within each sub-graph, and determining a risk score for each blockchain account address and smart contract address associated with each sub-graph.

Methods and systems for forensic investigations in contract networks

A method for generating a merged large language model (LLM) from input data including receiving derived prompts generated from a user prompt and relevant contexts, receiving the relevant contexts related to the generation of the plurality of derived prompts, providing derived prompts and the relevant contexts to one or more merged LLMs, generating a plurality of results, and sending the plurality of results to an output broker.

A system for adaptive LLM processing including an AI broker that receives an input query, generates derived queries from the input query, transmits the derived queries to LLMs, receives results from the plurality of LLMs, maintains an AI broker database of historical query paths for received input queries and corresponding stored results, evaluates an effectiveness of query paths based on the stored results, adjusts generation of subsequent derived queries based on the evaluated effectiveness, and iteratively refines generating derived queries until a predetermined service level assurance is achieved. The AI broker includes an input broker component to generate the derived queries and an output broker component to process the results, assess a quality of the results received from the plurality of LLMs, store successful query paths and corresponding results in the AI broker database, and modify future query processing based on stored successful query paths.

Systems and methods of answering queries using one or more h-LLMs including receiving user prompts via an API at a query layer from a user interface, determining an analysis mode of the user prompts to be a real-time mode, transmitting a query prompt including the user prompts from the query layer to a real-time layer responsive to determining the analysis mode being a real-time mode, the real-time layer including an h-LLM and being capable of generating a result in a response period on the order of seconds, receiving a real-time response from the real-time layer responsive to the query prompt, and transmitting the real-time response to the user interface via the API.

A method of processing language model input data in a distributed computing environment, including receiving an input data stream; tokenizing data received from the input data stream into a plurality of tokens using a map-reduce operation in the distributed computing environment, processing the plurality of tokens in parallel in the distributed computing environment to produce a plurality of processed tokens, aggregating the plurality of processed tokens using a reduce operation to produce a plurality of aggregated tokens, generating one or more updated incrementally-updated family of large language models (h-LLMs) by updating one or more incrementally-updated h-LLMs with the plurality of aggregated tokens in real-time, and responding to a user query using the one or more updated incrementally-updated h-LLMs.

A method of generating outputs in LLMs including receiving including textual content, defining a context for the documents including identifying a topic or a category, segmenting the textual content into content chunks associated with the topic or category, assigning a tag to each content chunk, identifying selected chunks, adding metadata to the selected chunks indexing the selected chunks, receiving a query, and performing a response generation process including determining if a cache includes information for the query and either retrieving the information from the cache or performing a search of the index to retrieve the information, generating an augmented query by augmenting the query with retrieved information, generating a response from the augmented query, evaluating the response for compliance with criteria, and one of generating a final response and transmitting the final response to the user or performing a fine-tuning process comprising redefining the of the one or more contexts.

Method and system for optimizing use of retrieval augmented generation pipelines in generative artificial intelligence applications

A system and method of improving performance of LLMs including receiving a query, retrieving relevant documents for the query to form a combined context, partitioning the combined context in context partitions, generating intermediate results from the context partitions using a mapper prompt, and generating a final result from the intermediate results using a reducer prompt. The final result is transmitted to a user.

Growing community of inventors

Chandigarh, India

Arshdeep Bahga