Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
Claims 1-20 are pending. Claims 1, 8, and 15 are independent.
This Application is not yet published.
Apparent priority: 17 March 2025.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(2) the claimed invention was described in a patent issued under section 151, or in an application for patent published or deemed published under section 122(b), in which the patent or application, as the case may be, names another inventor and was effectively filed before the effective filing date of the claimed invention.
Claims 1-2, 7-9, and 14-16 are rejected under 35 U.S.C. 102(a)(2) as being anticipated by Raghuvanshi (US 20260135772).
PNG
media_image1.png
343
510
media_image1.png
Greyscale
Regarding Claim 1, Raghuvanshi teaches:
1. A method for utilizing a language model [Raghuvanshi. “[0035] … In one example, MLMs of the present disclosure may include an ML-based generative model, such as a language model, e.g., a “large language model” (LLM). For instance, an ML-based generative model used in the present examples may comprise a generative adversarial network (GAN), a bidirectional encoder representations from transformers (BERT) model (e.g., BERT-Base, BERT-Large, etc.), a generative pre-training (GPT) model (e.g. GPT, GPT-2, GPT-3, or the like), a semantic graphs-based pre-training (SGPT) model, or other generative natural language processing (NLP) models. In one example, MLMs of the present disclosure may comprise an ada text embedding model.”] to implement a configuration change in a network, [Raghuvanshi. “… The processing system may then reconfigure the communication network according to the network configuration.” Abstract.]
the method comprising:
deploying the language model to a network controller that is configured to respond to inputs from network administrators associated with the network; [Raghuvanshi, Figure 2, “client 291” teaches the “network administrators” because it is the client that sets policy. “Gen AI Engine 292” including the “generative MLM 293” teaches the “network controller” that responds to client intent. “A processing system including at least one processor may obtain a client intent for a network service in a communication network. …” Abstract.]
receiving, by the language model, an input from a network administrator indicating a description of a requirement for a configuration change; [Raghuvanshi, , Figure 2, “client 291”/administrator is sending a “network service request 201” to the MLM 293/ Language Model. Figure 3, 310. “[0056] At optional step 310, the processing system may obtain a network service request, e.g., in a natural language format, from the client system. For instance, the network service request may be for a network service in a communication network. Example network service requests may include “I'd like a new virtual private network for our enterprise locations in New York and California,” “I'd like support massive multiplayer online gaming using my new VR headset,” or the like.” Steps 325 and 330 show that the intent is for reconfiguration of the network. Figure 2, 207: “Provide Network configuration to 5Mo…” See [0061] and [0062] discussing the reconfiguration/ configuration change in detail.]
determining, by the language model, a series of actions to execute to implement the configuration change; and [Raghuvanshi, Figure 2, the network configuration provided by the MLM/LLM includes functions 208, 209, 210, 211 to end in verifying the deployment at 212. “[0051] At 206, the generative AI engine 292 may obtain a network configuration as the output of the generative MLM 293 in response to the intent and network standards documentation as input vector/prompt and supplemental prompt content, respectively. The network configuration may include one or more network functions to support the network service, one or more configuration settings of the one or more network functions to support the network service, the desired links/interfaces and bandwidth for the links/interfaces, etc. In the present example, the request (and the client intent) may be for a low latency network service. In this case, the recommended network configuration may include the four network elements, or network functions (NFs): AMF, SMF, UPF, and gNB. At 207, the generative AI engine 292 may provide the network configuration to SMO 295, e.g., with a request/instruction(s) to implement the network configuration. SMO 295 may then deploy the respective NFs at 208-211, respectively….”]
outputting, by the language model, the series of actions to execute to the network administrator. [Raghuvanshi, Figure 2, 206: Obtain network configuration from the MLM 293. Figure 2, 207 and Figure 3, 325 and 330. “[0062] At step 330, the processing system reconfigures the communication network according to the network configuration. For example, the reconfiguring may include instantiating one or more network functions via NFVI (e.g., host devices, SDN nodes, or the like) to support the network service, transmitting at least one instruction to at least one NF to adjust at least one configuration setting to support the network service, or allocating bandwidth on at least one interface to support the network service. In one example, the reconfiguring may include transmitting an instruction to a service management orchestrator (SMO) to implement the reconfiguring according to the network configuration….”]
Regarding Claim 2, Raghuvanshi teaches:
2. The method of claim 1, wherein the language model is trained on network documentation and verbiage. [Raghuvanshi teaches that the network documentation can help the LLM reach a result when supplied as context for the prompt and then teaches that these past prompts that include the documentation/policies/ constraints can be used for the retraining of the LM. Figure 2, “Documentation Repository 294.” “14. The method of claim 1, further comprising: obtaining network standards documentation associated with the client intent, wherein the network standards documentation is applied as supplemental prompt content to the generative machine learning model along with the input vector.” It also teaches: “[0011] … Examples of the present disclosure may further apply the client intent along with the network architecture standards documentation (e.g., as supplemental prompt content) to a generative machine learning model (MLM) that is trained to generate a recommended network configuration or “network blueprint” (e.g., that is expected to support the client intent) as an output. For instance, network blueprint may include one or more network functions (NFs) to support the client intent, e.g., a requested network service, as well as interfaces between NFs, micro services, configurations/settings, etc. on the NFs, or the like.” “[0035] It should be noted that as referred to herein, a machine learning model (MLM) (or machine learning-based model) may comprise a machine learning algorithm (MLA) that has been “trained” or configured in accordance with input training data to perform a particular service….” “[0068] … In one example, the method 300 may include storing NL requests, the prompts/input vectors used, the network configurations generated therefrom, etc. e.g., for ongoing learning and model retraining, etc. For instance, in one example, the method 300 may include obtaining feedback and retraining the intent mapping function and/or the generative model(s)….”]
Regarding Claim 7, Raghuvanshi teaches:
7. The method of claim 1, wherein the language model is a large language model (LLM). [Raghuvanshi, “[0035] … In one example, MLMs of the present disclosure may include an ML-based generative model, such as a language model, e.g., a “large language model” (LLM)….”]
Claim 8 is a system claim with limitations corresponding to the limitations of Claim 1 and is rejected under similar rationale. Additionally, Raghuvanshi, Figure 4, teaches “processors 402” and “memory 404” and the “storage device 406.”
Claim 9 is a system claim with limitations corresponding to the limitations of Claim 2 and is rejected under similar rationale.
Claim 14 is a system claim with limitations corresponding to the limitations of Claim 7 and is rejected under similar rationale.
Claim 15 is a computer program product system claim with limitations corresponding to the limitations of method Claim 1 and is rejected under similar rationale. Additionally, Raghuvanshi, Figure 4, teaches “processors 402” and “memory 404” and the “storage device 406.”
Claim 16 is a computer program product system claim with limitations corresponding to the limitations of method Claim 2 and is rejected under similar rationale.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 3-4, 10-11 and 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Raghuvanshi in view of Yan (U.S. 20260172319).
Regarding Claim 3, Raghuvanshi teaches:
3. The method of claim 1, wherein the language model is a first language model and [Raghuvanshi, Figure 2, “Generative MLM 293.”]
wherein determining the series of actions further comprises:
determining a series of subtasks for the configuration change; [Raghuvanshi, Figure 2, 207 provides a series of functions/tasks 208, 209, 210, 211, and 212 each of which becomes a subtask of 207. Figure 3, 330. “[0051] At 206, the generative AI engine 292 may obtain a network configuration as the output of the generative MLM 293 in response to the intent and network standards documentation as input vector/prompt and supplemental prompt content, respectively. The network configuration may include one or more network functions to support the network service, one or more configuration settings of the one or more network functions to support the network service, the desired links/interfaces and bandwidth for the links/interfaces, etc. In the present example, the request (and the client intent) may be for a low latency network service. In this case, the recommended network configuration may include the four network elements, or network functions (NFs): AMF, SMF, UPF, and gNB. At 207, the generative AI engine 292 may provide the network configuration to SMO 295, e.g., with a request/instruction(s) to implement the network configuration….”]
inputting a description of each subtask into a second language model; and [Raghuvanshi mentions having different LMs: “[0050] In one example, different MLMs may be possessed by the generative AI engine 292, where based on the accuracy/quality of the response/output, these MLMs can be reconfigured/retrained in an adaptive way. As such, in one example, the generative MLM 293 may comprise one or more MLMs that are selected via an auto-ML process….” Raghuvanshi also teaches having different functions/ tasks/ substasks once determined by the original MLM/LM are performed. The two teachings are conducive to having different MLMs for different sub-tasks but this is not taught by the reference.]
receiving from the second language model an action to execute for each subtask. [Raghuvanshi does not include a second LM but the actions/functions are executed once output by the first MLM/LM.]
Raghuvanshi teaches the one MLM 293 that generates the subtasks 208-212 and teaches that different MLMs may be fine tuned for different tasks. Does not expressly teach having different MLMs for the sub-tasks.
Yan teaches:
wherein determining the series of actions further comprises:
determining a series of subtasks for the configuration change; [Yan: the sub-tasks are related to network functions (NF) each of which gets a different command and the first LM determines the commands for each sub-task which are then input to the pertinent sub-task: “[0030] For instance, the natural language request for the cellular network management task may comprise a request for available commands to configure at least one aspect of the cellular network, where an output may comprise the available command(s). This task may involve the language model agent first determining the relevant portion of the cellular network, e.g., base station equipment in a tracking area, core NFs associated with a particular network slice, etc. For instance, this may include querying a network topology database to identify the relevant network elements and/or network functions. A next sub-task may include querying a network status database to determine a current state of this portion of the network, e.g., the current configurations (e.g., current configuration setting values) and/or current network performance indicators (e.g., indicating current load, utilization, and/or availability levels of NFs and/or components thereof (e.g., current processor capacity, memory capacity, etc.), bandwidth availability and/or utilization, alarm data, etc.). Then, given the current state of the network (or at least the portion thereof), different commands may be available. For instance, if a gNB is in a sleep mode or other modes (e.g., a power saving state, a reduced function mode, etc.) in which antennas are off, it may not be a valid command to reduce the number of antennas in use for power saving. However, valid commands may include a command to change the power state (e.g., to a lower power state or to a higher power state).” See [0028] also. “[0028] … or instance, this functionality may be represented by a planning module of a language model agent. Notably, server(s) 135 may decompose a fulfillment of a natural language request performing a cellular network management task into a sequence of one or more sub-tasks (e.g., to be performed sequentially and/or in parallel). In one example, server(s) 135 may maintain a state of a task flow, e.g., as the one or more sub-tasks are processed and completed….”]
inputting a description of each subtask into a second language model; and [Yan: each NL command may go to its own LM: “[0031] In one example, different language models may be used in the language model core to perform sub-tasks. For instance, a first language model may be used to determine that a first agent tool is to be used to obtain at least one dynamic data component and that a second agent tool is to be used to obtain at least one static data component. A different language model may be used to determine how to handle an output after gathering the at least one dynamic data component and the at least one static data component. …” ]
receiving from the second language model an action to execute for each subtask. [Yan: the language models of Yan take in natural language inputs and generate commands that execute the tasks: “[0031] … For instance, to retrieve data from a database, an SQL query may be used, where an agent tool may be tasked with formulating a proper SQL query, e.g., given an input of a natural language request or a high level intent for retrieval of data having particular characteristics as indicated in the natural language request or intent. However, in another example, the prompt from the planning module to the language model core may include sufficient information about the agent tool (e.g., detailed information about a database query API) such that the language model core may determine that the use of the database query agent tool is warranted and such that the language model core may also formulate the query itself.”]
Raghuvanshi and Yan pertain to network reconfiguration using LMs and it would have been obvious to combine the different LMs of Yan each of which is trained for a sub-task with the system of Raghuvanshi which does mention a set of LMs instead of one in order to improve the functioning of the overall system. This combination falls under combining prior art elements according to known methods to yield predictable results or use of known technique to improve similar devices (methods, or products) in the same way. See MPEP 2141, KSR, 550 U.S. at 418, 82 USPQ2d at 1396.
Regarding Claim 4, Raghuvanshi, teaches having different LMs and fine tuning of its LMs for different tasks: “[0050] In one example, different MLMs may be possessed by the generative AI engine 292, where based on the accuracy/quality of the response/output, these MLMs can be reconfigured/retrained in an adaptive way. As such, in one example, the generative MLM 293 may comprise one or more MLMs that are selected via an auto-ML process. For instance, an operator may provide one or more optimization criteria to obtain the best performing model(s) with respect to accuracy, speed, a combination of such factors, etc. In addition, in one example, the generative MLM 293 may be adapted from a pre-trained model, where the framework of the generative MLM-based communication network knowledge platform may be used to modify and retune the adopted model(s), e.g., specifically to generate recommended network configurations, and in one example, more specifically with respect to a particular network operator's communication network. Thus, it should be noted that training of the generative MLM 293 can be accomplished in different ways such as training from scratch, fine-tuning of a pre-trained model, retrieval-augmented generation (RAG), reinforcement learning using feedback, prompting/prompt-tuning, learning using adapters, a combination of any of the foregoing, and so forth.” This teaching suggests the subject of Claim 4.
Yan teaches:
4. The method of claim 3, wherein the second language model is fine-tuned with datasets that include one or more actions to execute for a description of a network configuration change. [Yan teaches decomposing the reconfiguration task into a set of subtasks and using fine-tuned LMs for each sub-task where the input to each LM is a natural language description of the sub-task.. “[0028] … Notably, server(s) 135 may decompose a fulfillment of a natural language request performing a cellular network management task into a sequence of one or more sub-tasks (e.g., to be performed sequentially and/or in parallel). In one example, server(s) 135 may maintain a state of a task flow, e.g., as the one or more sub-tasks are processed and completed. In one example, the state of the sequence/task flow as well as intermediate data, such as dynamic and static data components retrieved at respective sub-tasks, may be stored in the memory unit of the server(s) 135….” “[0031] In one example, different language models may be used in the language model core to perform sub-tasks. For instance, a first language model may be used to determine that a first agent tool is to be used to obtain at least one dynamic data component and that a second agent tool is to be used to obtain at least one static data component. A different language model may be used to determine how to handle an output after gathering the at least one dynamic data component and the at least one static data component…. For instance, to retrieve data from a database, an SQL query may be used, where an agent tool may be tasked with formulating a proper SQL query, e.g., given an input of a natural language request or a high level intent for retrieval of data having particular characteristics as indicated in the natural language request or intent….”]
Rationale similar to that provided for Claim 3.
Claim 10 is a system claim with limitations corresponding to the limitations of Claim 3 and is rejected under similar rationale.
Claim 11 is a system claim with limitations corresponding to the limitations of Claim 4 and is rejected under similar rationale.
Claim 17 is a computer program product system claim with limitations corresponding to the limitations of method Claim 3 and is rejected under similar rationale.
Claim 18 is a computer program product system claim with limitations corresponding to the limitations of method Claim 4 and is rejected under similar rationale.
Claims 5-6, 12-13 and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Raghuvanshi and Yan further in view of Hensley (US 20250251932).
Regarding Claim 5, Raghuvanshi includes a self-optimizing (SON) network orchestrator to activate and deactivate network components. [0023]. It also teaches that its MLMs are trained in an adaptive way: “[0050] In one example, different MLMs may be possessed by the generative AI engine 292, where based on the accuracy/quality of the response/output, these MLMs can be reconfigured/retrained in an adaptive way…” This means they are optimized.
Yan hints that its method is to optimize the network. [0009].
Neither sets forth the steps of the following Claim which is a particular method of training of LMs.
Hensley teaches:
5. The method of claim 3, wherein each subtask is input into the second language model multiple times and further comprising:
generating, by the second language model, multiple possible actions to execute to implement the subtask; [Hensley teaches the steps of this Claim which is directed to a particular optimization method called GRPO: “[0077] Group Relative Policy Optimization (GRPO) may be used to train the reasoning model to improve reasoning capabilities efficiently while reducing computational costs. Unlike some policy optimization methods that require a separate critic model, GRPO estimates the training baseline from grouped scores, reducing resource overhead. During training, for each input query, GRPO samples multiple candidate outputs from the current policy. These outputs may be evaluated using a reward function, which in some cases, includes accuracy-based and format-based rewards. The model's objective may be to maximize the probability of higher-reward outputs while penalizing deviations from prior behavior using a Kullback-Leibler (KL) divergence term to maintain training stability. In some embodiments, the optimization process follows a clipped policy gradient update, ensuring that updates are within a bounded range to prevent overly aggressive changes. By leveraging group-based relative comparisons instead of absolute value-based critiques, GRPO is expected to afford more stable and efficient reinforcement learning, facilitating the emergence of complex reasoning behaviors.”]
evaluating, by the second language model, each action of the multiple possible actions; and [Hensley, “[0077] … These outputs may be evaluated using a reward function, which in some cases, includes accuracy-based and format-based rewards….”]
determining, by the second language model, an optimal action to execute from the multiple possible actions. [Hensley, “[0077] … The model's objective may be to maximize the probability of higher-reward outputs while penalizing deviations from prior behavior using a Kullback-Leibler (KL) divergence term to maintain training stability. In some embodiments, the optimization process follows a clipped policy gradient update, ensuring that updates are within a bounded range to prevent overly aggressive changes. ….”]
Raghuvanshi/Yan ang Hensley are directed to optimizing LMs and it would have been obvious to use the particular optimization method of Hensley with the system of combination as one known choice. This combination falls under simple substitution of one known element for another to obtain predictable results or use of known technique to improve similar devices (methods, or products) in the same way. See MPEP 2141, KSR, 550 U.S. at 418, 82 USPQ2d at 1396.
(See also Lo in the Conclusion.)
Regarding Claim 6, Raghuvanshi teaches:
6. The method of claim 5, wherein the optimal action is customizable and further based at least in part on policy requirements of an organization. [Raghuvanshi, Figure 2 shows that the “Documentation Repository 294” is providing network standards documentation (NSD) to (203) and from (204) MLM 293. NSD teaches the policy requirements of the organization. “[0046] In 203, the generative AI engine 292 may retrieve network standards documentation from a documentation repository 294 (such as DB(s) 136 in FIG. 1, or the like). In one example, the network standards documentation may be 3GPP technical standards and/or others such as mentioned above. In one example, the network standards documentation may be specific to the client intent determined at 202. For instance, the generative AI engine 292 may obtain the top “N” number of documents that may be relevant to the client intent….” “[0047] In any case, the generative AI engine 292 may obtain the network standards documentation at 204 and at 205 may apply the client intent and the network standards documentation to the generative MLM 293. For example, the client intent may comprise an input vector/prompt for the generative MLM 293. In addition, the network standards documentation may be applied as supplemental prompt content….” “[0045] … For instance, new client intents may be defined by different sets/combinations of network performance indicator thresholds and/or other requirements/constraints….” “[0048] In one example, client information may be further provided as supplemental prompt content. For instance, it may be learned over time that a client may object to all network deployments/network configurations that involve NFs that reside on anon-domestic (e.g., non-United States) physical infrastructure, e.g., due to specific contract requirements and/or confidentiality concerns….”]
Claim 12 is a system claim with limitations corresponding to the limitations of Claim 5 and is rejected under similar rationale.
Claim 13 is a system claim with limitations corresponding to the limitations of Claim 6 and is rejected under similar rationale.
Claim 19 is a computer program product system claim with limitations corresponding to the limitations of method Claim 5 and is rejected under similar rationale.
Claim 20 is a computer program product system claim with limitations corresponding to the limitations of method Claim 6 and is rejected under similar rationale.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Kumar (U.S. 20260189599):
Kumar is directed to “COMMUNICATION NETWORK MANAGEMENT SYSTEM WITH FEDERATED LEARNING” and uses Machine Learning Models (MLM) which could be LLMs: “[0037] In one example, MLMs of the present disclosure may include an ML-based generative model, such as a language model, e.g., a “large language model” (LLM). For instance, an ML-based generative model used in the present examples may comprise a generative adversarial network (GAN), a bidirectional encoder representations from transformers (BERT) model (e.g., BERT-Base, BERT-Large, etc.), a generative pre-training (GPT) model (e.g. GPT, GPT-2, GPT-3, or the like), a semantic graphs-based pre-training (SGPT) model, or other generative natural language processing (NLP) models. In one example, MLMs of the present disclosure may comprise an ada text embedding model.”
Kumar. Figure 3 begins with the training of the model. “[0044] FIG. 3 illustrates a flowchart of an example method 300 for tuning an aggregated machine learning model in accordance with first parameters of a first machine learning model relating to a first network zone and second parameters of a second machine learning model relating to a second network zone, in accordance with the present disclosure. …” Figure 3, 310, 315, 320, 325 pertain to training and fine tuning of the trained models. [0045]-[0046] discuss the type of data used for the training of the models: “[0045] At step 310, the processing system obtains first data samples relating to a first network zone of a communication network and second data samples relating to a second zone of the communication network….” “[0046] In addition, the first data samples may comprise first network status information and the second data samples comprise second network status information, e.g., for the respective network zones. For instance, as noted above, network status information may include measured performance indicators such as peak and average processor utilization, average memory utilization, bandwidth utilization, or the like, packet loss rate, call failure rate, call drop rate, packet delay, packet throughput, jitter, signal to noise (SNR) ratio, measurements of a video uplink data rate and/or measurements of a video downlink data rate, VMAF metrics, and so forth. In accordance with the present disclosure, network status information may also include network configuration settings for various network elements/network functions, such as setting values for various configurable settings and/or a network topology, which may be indicated by the setting values, or otherwise.”
Kumar, Figure 3, 320: “[0048] At step 320, the processing system trains a second MLM for a second prediction task associated with the second network zone using the second data samples, where the first prediction task and the second prediction task are both of a same first type of prediction task….”
Lo (U.S. 20260227970):
PNG
media_image2.png
638
796
media_image2.png
Greyscale
Schrader (U.S. 20250110786):
[0085] As noted above, the agent system 102 (e.g., the agent service 106) may include and/or have access to one or more large language model (LLM) or other language model, and the LLM may be fine-tuned or trained on appropriate training data (e.g., annotated data showing correct or incorrect pairings of sample natural language queries and responses). After receiving a user input, the agent system 102 may generate and provide, through the agent service 106, one or more prompts to a LLM 130a, which may include one or more large language models trained to fulfill a modeling objective, such as task completion, text generation, summarization, or the like.
PNG
media_image3.png
674
524
media_image3.png
Greyscale
Castrejon Subira (U.S. 20250239066):
PNG
media_image4.png
320
538
media_image4.png
Greyscale
Bhat (U.S. 20250383970):
PNG
media_image5.png
366
528
media_image5.png
Greyscale
Any inquiry concerning this communication or earlier communications from the examiner should be directed to FARIBA SIRJANI whose telephone number is (571)270-1499. The examiner can normally be reached 9 to 5, M-F.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Pierre Desir can be reached at 571-272-7799. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/Fariba Sirjani/
Primary Examiner, Art Unit 2659
Prosecution Insights