DETAILED ACTION
Claims 1-20 rejected under 35 USC § 101.
Claims 1-20 rejected under 35 USC § 103.
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 .
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Claims 1 and 11
Step 2A, Prong One. Claims 1 and 11 recite receiving a computational workflow specification containing stages with respective objectives, exit criteria, and fillable fields, together with a directed graph arranging the stages, and computing filled field values based on that specification. Determining the content to place in the unfilled portions of a specified procedure encompasses evaluating the supplied requirements and composing instructions that implement them. For example, a person can read a two stage specification requiring a form completeness check followed by a request for missing information, and write instructions or a short prompt into each corresponding blank. The objectives describe the intended work, the exit criteria describe when to proceed, and the directed graph describes the relationships among the stages. Those informational requirements do not prevent the content determination from being practically performed by a person with pen and paper. This evaluation and composition falls within the mental process grouping. MPEP § 2106.04(a)(2), subsection III.
The generative language model implements the content determination. Its recitation does not change the claimed content determination into a mathematical exception, the claims specify no mathematical calculation. The processor and model implementation, and the subsequent computer execution of the completed workflow, are additional elements considered below. The identified mental process is determining the missing procedural content, not performing neural network inference or executing computer instructions in the human mind.
Step 2A, Prong Two. This judicial exception is not integrated into a practical application. Claim 1 adds a computing system with processing devices, and claim 11 performs the method for use with a computing system. Both claims require the field values to e computed at least in part at a generative language model and require execution of the resulting workflow. The computing implementation thus receives the procedure description, uses the model to supply missing content, and operates the resulting procedure.
The directed graph and the association of an objective, exit criterion, and fillable field with each relevant stage organize the input to the content generation operation. The claims require generation based at least in part on the workflow specification, but do not specify a procedure for translating the respective objectives and exit criteria into a n implementation, validating the implementation against those criteria, or changing the operation of the model. The graph does not require a particular execution control arrangement beyond arranging the stages. The model is invoked for its content generation function with a claimed change to that function.
Executing the filled workflow is an actual computer operation. As recited, however, it requires running whatever computational procedure results from the content generation. The claims do not identify a particular technical task performed by that procedure or a particular mechanism by which running it improves computer functionality. The combination therefore supplies and executes the procedural content through generally recited computer functions. The sequence of receiving the specified plan, generating its missing implementation content, and running the completed procedure does not impost a further technological limitation on the identified content determination. The additional elements amount to instructions to implement that determination and use its result on a computing system. MPEP §§ 2106.04(d)(1), 2106.05(a), and 2106.05(f).
Step 2B. The claims do not include additional elements sufficient to amount to significantly more than the judicial exception. The computing system receives the specification, invokes a generative language model to supply field contents, and executes the resulting workflow. To the extent filling fields includes retaining those values as electronic records, that recordkeeping function corresponding to the generic function identified in MPEP § 2106.05(d), subsection II, computer functions item iii, “Electronic recordkeeping.” This finding concerns retaining the value, rather than their generation by the model.
The model based generation and workflow are evaluated under MPEP § 2106.05(f). The model supplies the procedural content that constitutes the identified mental activity, and execution uses that content to perform an otherwise unspecified computational task. Considering the elements as an ordered combination, the claim prescribes the input representation, the desired completion of that representation, and use of the completed result. It supplies no additional technique for achieving a technological improvement beyond implementing the content determination and running its result. The additional elements consequently do not amount to significantly more. MPEP §§ 2106.05 and 2106.05(f).
Claims 2 and 12
Claims 2 and 12 further recite that at least one filled value is executable code. Code supplies the instructions implementing the procedure, and a person can compose a short code fragment in response to the specified requirements. The inherited system uses the generative language model to perform that composition and then executed the filled workflow. The claims add no particular code generation, compilation, validation, or execution technique. Together with the inherited limitations, identifying the generated content as executable code does not integrate the mental process into a practical application or supply significantly more. MPEP §§ 2106.04(a)(2), 2106.04(d)(1), and 2106.05(f).
Claims 3 and 13
Claims 3 and 13 further recite that at least one filled value is a prompt of a machine learning model called in the corresponding filled stage. The prompt supplies instructions to the called model, placing the generated content in use during the workflow. The claim does not specify a prompt construction technique or a relationship between prompt structure and model operation that produces a particular technical improvement. Composing an instruction for the called model and using that instruction in its corresponding stage implements the inherited procedural content determination. Together with the inherited limitations, there features do not integrate the exception into a practical application or supply significantly more. MPEP §§ 2106.04(d)(1) and 2106.05(f).
Claims 4 and 14
Claims 4 and 14 further recite multiple workflow generation iterations and, in at least one iteration, receiving an unfilled stage through a user interface and outputting a filled stage to that interface. These limitations permit the procedural content to be supplies and reviewed in successive portions. The interface receives the information used in generation and presents the generated result; the claims specify no interface arrangement or interaction mechanism that changes computer operation. Repeating the generation procedure and exchanging its input and results through the interface does not add a technological implementation of the underlying evaluation and composition. Together with the inherited limitations, these features do not integrate the exception into a practical application or supply significantly more. MPEP §§ 2106.05(f) and 2106.05(g).
Claims 5 and 15
Claims 5 and 15 further recite, within the iterations inherited from claims 4 and 14 respectively, receiving a user revision to a filled value and modifying that value as specified. This adds editing of the generated procedural content. The revision identifies the desired content change, and the system applies that change without a further claimed editing, consistency checking, or validation mechanism. Retaining the revised value is electronic recordkeeping, and making the requested revision carriers out the users content determination. Together with the inherited iterative interface and generation limitations, these features do not integrate the exception into a practical application or supply significantly more. MPEP §§ 2106.05(d).
Claims 6 and 16
Claims 6 and 16 further recite performing “retrieval-augmented generation (RAG) based at least in part on the stage objective or exit criterion, obtaining a RAG query result, and using that result to compute the filled value. This connects the retrieval to the particular stage and makes the retrieved information as input to generation. The added operation obtains information relevant to the procedural requirements and uses that information to supply the missing content. The claim prescribe neither an improved retrieval or indexing technique nor a particular context management mechanism; the query subject matter and the use of its result define what information informs generation. The technical RAG implementation performs retrieval and generation in those states roles. Considered with he inherited graph, field representation, and execution, this information retrieval supplies material for the same procedural content determination without adding a particular technological solution. It therefore does not integrate the exception into a practical application or supply significantly more. MPEP §§ 2106.04(d)(1), 2106.05(f), and 2106.05(g).
Claims 7 and 17
Claims 7 and 17 further recites that “the directed graph structure includes at least one cycle pattern.” This limitation constrains the arrangement of the stage by providing a path returning to a prior stage. In the inherited workflow specification, it describes a possible repeated portion of the procedure whose missing content is determined and which is then executed. The claims do not specify a relationship between traversal of that path and generation of the field values, satisfaction of an exit criterion, or correction of stage output. The cycle therefore does not require the recursive conversational interaction described in the specification. Even when the cycle is traversed during execution, the added limitation requires repetition within the procedure without specifying a control or feedback operation that improves its computer implementation. The processing devices and generative language model continues to supply and execute the procedural based content, now arranged to permit repetition. Considered individually and in the ordered combination with the stage objectives, exit criteria, fillable fields, model based generation, and workflow execution, the cyclic relationship limits the procedure’s logical sequence but adds no implementation beyond generating and executing that procedure. It therefore neither integrates the inherited mental process into a practical application nor supplies significantly more than the exception. MPEP §§ 2106.04(d)(1), 2106.05(a), and 2106.05(f).
Claims 8 and 18
Claims 8 and 18 further recite that the directed graph includes “a fan-in pattern, a fan-out pattern, and/or a map-reduce pattern.” These patterns constrain the relationships among the workflow portions by permitting inputs to converge, outputs to diverge, or outputs to be combined. In the inherited combination, they organize the procedure whose missing content is generated and which is then executed. The claims do not prescribe a scheduling, synchronization, partitioning, aggregation, or resource management technique for implementing those relationships. Because the alternatives do not require all three patterns together, the claims also encompass a workflow having only one branching or converging pattern. These additional logical relationships do not change the generally recited manner of generating and executing the procedural content. Together with the inherited limitations, they do not integrate the exception into a practical application or supply significantly more. MPEP §§ 2106.04(d)(1), 2106.05(a), and 2106.05(f).
Claim 9
Claim 9 further recites that, “during execution of the filled computational workflow,” the processing devices are configured to “dynamically select an input forma and/or an output format of at least one workflow stage of the plurality of workflow stages.” Choosing a format as the procedure proceeds encompasses a person deciding whether the information currently being proceeded should be produced as text or an image. The claim specified neither a scale nor a selection criterion that prevents this choice from practically being performed in the human mind. That choice is an evaluation within the mental processes grouping. The processing devices implement the choice during actual workflow execution, but the claim does not specify a format conversion, interface negotiation, or compatibility operation that gives the selection a particular technological application. The output format alternative does not require adaptation of an interface between two computational stages. With the inherited specification, graph, generative language model, and filled workflow, the limitation requires the system to supply procedural content, execute it, and choose an input or output format while it runs. The ordered combination implements those content and format choices through the stated computing functions without adding a technique for improving how the computer represents, converts, or processes the information. Together with the inherited limitations, the added feature does not integrate the mental processes into a practical application or supply significantly more. MPEP §§ 2106.04(a)(2), 2106.04(d)(1), and 2106.05(f).
Claims 10 and 19
Claims 10 and 19 further recite conditional routing of a workflow stage output during execution of the filled computational workflow. Under the definition in paragraph 0040 of the specification, conditional routing selects a subsequent stage from a plurality of stages according to one or more properties of the current stage output. This selection encompasses observing that an output is text or an image and choosing a corresponding text processing or image processing stage, as illustrated in paragraph 0041. Evaluating that property and choosing the next stage can practically be performed in the human mind and falls within the mental processes grouping. The computing implementation directs the output to the selected stage during execution, but the claims do not specify how routing is implemented beyond that conditional selection. In particular, they do not require the distribution of routing instructions among stages that produces the reduced communication and latency described in paragraph 0043. Considered with the inherited generation of field values from stage objectives and exit criteria, the graph, and execution of the filled workflow, the added limitation makes the procedure’s next step depend on the evaluated output property. The generative language model supplies procedural content and the computing system carries out the resulting choices; their ordered combination supplies no further routing or execution technique beyond implementing those choices. Together with the inherited limitations, conditional routing therefore does not integrate the mental processes into a practical application or supply significantly more. MPEP §§ 2106.04(a)(2), 2106.05(a), and 2106.05(f).
Claim 20
Step 2A, Prong One. Claim 20 recites the same receipt, field completion, and execution arrangement as claim 1, with a plurality of unfilled stages and filled values including both executable code and a prompt of a machine learning model called in the corresponding stage. Composing instructions for multiple portions of a procedure, including writing a short code fragment and a prompts used by another portion, encompasses the same evaluation and composition performed by a human programmer. The claims does not require a scale of stages or content that prevents that determination from being practically performed with pen and paper. The additional content types do not remove the recited mental process. MPEP § 2106.04(a)(2), subsection III.
Step 2A, Prong Two. This judicial exception is not integrated into a practical application. The processing devices and generative language model supply the code and prompt values, and the filled workflow is executed with a model call in the corresponding stage. The claim specifies both types of generated content but does not specify a technical coordination mechanism between them, a validation or compilation technique, or an improvement to ether the generating model or the called model. Their placement in the stage fields supplies implementation content for the specified procedure. Considered together with the directed graph, stage objectives, exit criteria, and plurality of unfilled stages, these limitations require implementing and running the completed procedure through the stated computing tools. They do not require components or operations that produce a particular technological improvement. MPEP §§ 2106.04(d)(1), 2106.05(a), and 2106.05(f).
Step 2B. The claim does not include additional elements sufficient to amount to significantly more than the judicial exception. The electronic retention of generated field values has the recordkeeping function discussed for claim 1. The generation of both code and a prompt, the model call, and execution of the workflow operate in their stated roles to implement the completed procedures. Their ordered combination does not add a technique beyond supplying those forms of procedural content and executing them through the computing system. Requiring multiple unfilled stages repeats the same completion operation without adding a different computational implementation.
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 1-5, 7, 10-15, 17, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Dvorak et al., U.S. PG-Publication No. 2025/0005328 A1 (hereinafter DVORAK), in view of Friddle, U.S. PG-Publication No. 2024/0403005 A1 (hereinafter FRIDDLE).
Claim 1
DVORAK discloses a computing system comprising: one or more processing devices. ¶ 0054: Composite Ai system 110 includes a model editor, composition editor, simulation testing manager, and deployment manager. ¶ 0064: Generated services are compiled into container images, deployed to the platform’s compute infrastructure, and run.
DVORAK discloses receive a computational workflow specification.¶ 0064: Service code 152 is generated from a configured composition 151 of models. ¶ 0063: The architect creates, edits, and saves a composition, then publishes it to prepare for compilation and deployment.
DVORAK discloses a workflow specification including: a plurality of workflow stages, the plurality of workflow stages including one or more … workflow stages. ¶ 0038: The graph contains multiple model processing portions connected through control nodes, and processing repeats through additional stages.
DVORAK discloses a plurality of workflow stages that each include: a respective workflow stage objective. ¶ 0041: The composition combines multiple models with “corresponding solvers, goals, and APIs” into a service. ¶ 0055: Models are selected and combined for “each part of subproblem of the input problem or task to be performed.”
DVORAK discloses a respective workflow stage exit criterion. ¶ 0034: Orchestration defines “conditions of received outputs and when or how to proceed in processing” and may require specific response properties before proceeding. ¶ 0036: Parallel model outputs are intersected so that “only outputs that are in fuzzy match are passed to the next stage.”
DVORAK discloses one or more fillable fields. ¶ 0049: Service code is generated by “filling out template code.” Further, “A code or service template exists for every type of composition node.”
DVORAK discloses a directed graph structure in which the plurality of workflow
stages are arranged. ¶ 0038: Figure 4 describes a directed acyclic graph connecting control nodes, models, and computation services through successive stages.
DVORAK discloses based at least in part on the computational workflow specification, compute a filled computational workflow including one or more filled workflow stages, wherein computing the filled computational workflow includes computing respective filled values of the one or more fillable fields. ¶ 0064: The configured composition is transformed into a service; generated service code produces the composition with its models implemented as code. ¶ 0049: Generation fills component template code or configured services using JSON configuration files, environment variables, or API calls.
DVORAK discloses execute the filled computational workflow. ¶ 0064: A configured composition can be “transformed into a service, deployed, and run.”
DVORAK does not expressly disclose the plurality of workflow stages including a plurality of unfilled workflow stages; and computing respective filled values of the one or more fillable fields at least in part at a generative language model.
FRIDDLE discloses receive a computational workflow specification including: a plurality of workflow stages, the plurality of workflow stages including one or more unfilled workflow stages that each include: a respective workflow stage objective, a respective workflow stage exit criterion; and one or more fillable fields. ¶ 0271: The generation prompt supplies the process statement, process plan, and a commented JSON structure, and requests components of each individual step. ¶ 0151: The process data object contains a prompt and configuration component and a code component. The object may be populated at each step. ¶ 0164: There may be any number of logical blocks, and “all will have these components;” some steps need not be fully described. ¶ 0173: Logical blocks may contain pseudo instruction and subsequently be populated with relevant database information.
FRIDDLE discloses computing the filled computational workflow includes computing respective filled values of the one or more fillable fields at least in part at a generative language model. ¶ 0271: The LLM produces the JSON structure for individual steps. Deterministic logic then constructs the steps and assembles generated prompt components into an executable format.
It would have been obvious to a person having ordain skill in the art before the effective filing date of the claimed invention to modify the template based service generation of DVORAK to incorporate the generation of individual stage prompts and code from a process plan taught by FRIDDLE. One of ordinary skill in the art would be motivated to integrate that generation into DVORAK, with a reasonable expectation of success, in order to reduce token cost, generation time, and construction errors by generating structured components of each step and assembling them deterministically, as taught by FRIDDLE at ¶ 0271.
Claim 2
DVORAK discloses wherein at least one of the filled values included in the filled computational workflow is executable code. ¶ 0049: Service code is generated by “filling out template code” for composition components. Node implementations may use Python, C++, or scripts. ¶ 0064: Generated service code is compiled into container images, deployed, and run. The configured composition is “transformed into a service, deployed, and run.”
Claim 3
FRIDDLE discloses wherein at least one of the filled values included in the filled computational workflow is a prompt of a machine learning model called in the corresponding filled workflow stage. ¶ 0271: The process plan is translated into “individual prompts and python functions needed at each step of the process.” The language model produces structured components that deterministic logic assembles into individual prompts. ¶ 0272: After the individual prompts are generated, the structure of prompt file 514 is produced, followed by construction of the corresponding functions. ¶ 0161: Prompt 503 of logical block 515 “is the prompt that will get sent to the configured model.” Its parameters may be modified before execution.
Claim 4
FRIDDLE discloses wherein: the one or more processing devices are configured to receive the computational workflow specification and generate the filled computational workflow over a plurality of workflow generating iterations. ¶ 0269: The system supports “iteratively generating the process plan” with user augmentation during its construction. ¶ 0271: Once the process plan is reliable, the system generates the prompts and Python functions needed at each step. The language model produces structured components, and deterministic logic constructs the steps.
FRIDDLE discloses in at least one of the workflow generating iterations, the one or more processing devices are configured to: receive an unfilled workflow stage of the one or more unfilled workflow stages via a user interface. ¶ 0267: The process plan outlines individual steps and assigns tools to those steps before their implementation details are generated. ¶ 0268: In the graphical user interface, the user may update the process plan, and the corrected plan is sent for processing.
FRIDDLE discloses output a filled workflow stage of the one or more filled workflow stages to the user interface. ¶ 0273: After construction of the functional process in the manual context, the generated files needed for execution are returned to the user for possible updates. ¶ 0281: The completed process is returned to the user and “presented in an editor view.” The developer can then improve segments of the code or prompt sequence.
Claim 5
FRIDDLE discloses wherein, in at least one of the workflow generating iterations, the one or more processing devices are further configured to: receive a user revision to the filled value of a fillable field. ¶ 0273: After constructing the functional process, the generated execution files are returned to the user, who may “make updates to them.” ¶ 0281: The completed process is presented in an editor. The developer may “provide additional edits,” interacting with the interpreter to improve segments of the code or prompt sequence.
FRIDDLE discloses modify the filled value as specified by the user revision. ¶ 0282: When satisfied with their changes, the user can run the process in the interpreter.
Claim 7
DVORAK discloses wherein the directed graph structure includes at least one cycle pattern. ¶ 0037: Control nodes sit before and after models and model processing nodes in the system graph. In some embodiments, “a directed graph may be used with cycles.” These cycles specify “a loop between control nodes” so that information can be processed multiple times by the same models.
Claim 10
FRIDDLE discloses wherein, during execution of the filled computational workflow, the one or more processing devices are configured to perform conditional routing of a respective output of at least one workflow stage of the plurality of workflow stages. ¶ 0163: The Next section tells the interpreter “which logical block to switch to next.” Specifically, “an entirely different logical block may be jumped to based on a determination that a model did not return the correct output. ¶ 0207: Each logical block executes code before and after its model prompt. That code updates a data object used to modify the prompt or the “output which is returned to the user or carried to the next step.” ¶ 0208: The interpreter iterates through the logical blocks using their natural ordering or “configured variations of the execution path” indicated in the Next portion. ¶ 0206: In automatic mode, “the newly constructed sequence is now sent to the execute portion of the interpreter” and divided into steps or logical blocks.
Claims 11-15 and 19
Claims 11-15 and 19 are rejected utilizing the rationale for claims 1-5 and 10; the claims are directed to a method performed by the system.
Claim 20
DVORAK discloses a computing system comprising: one or more processing devices. ¶ 0054: Composite AI system 110 includes a model editor, composition editor, simulation testing manager, and deployment manager. ¶ 0064: Generated services are compiled into container images, deployed to the platform’s compute infrastructure, and run.
DVORAK discloses receive a computational workflow specification. ¶ 0064: Service code 152 is generated from configured composition 151. ¶ 0063: The architect edits and saves the composition, then publishes it to prepare compilation and deployment.
DVORAK discloses a workflow specification including: a plurality of workflow stages, the plurality of workflow stages including a plurality of … workflow stages. ¶ 0038: The graph contains multiple model processing portions and associated control nodes, with processing repeated through additional stages.
DVORAK discloses a plurality of workflow stages that each include: a respective workflow stage objective. ¶ 0041: The graph combines multiple models with “corresponding solvers, goals, and APIs.” ¶ 0055: Models are selected for “each part or subproblem of the input problem or task to be performed.”
DVORAK discloses a respective workflow stage exit criterion. ¶ 0034: Orchestration defines “conditions of received outputs and when or how to proceed in processing,” and may require specified response properties before proceeding. ¶ 0036: Outputs of parallel models are intersected so that “only outputs that are in fuzzy match are passed to the next stage.”
DVORAK discloses one or more fillable fields. ¶ 0049: Service code is generated by “filling out template code.” Further, “A code or service template exists for every type of composition node.”
DVORAK discloses a directed graph structure in which the plurality of workflow stages are arranged. ¶ 0038: Figure 4 describes a directed acyclic graph connecting control nodes, models, and computation services through successive stages.
DVORAK discloses based at least in part on the computational workflow specification, compute a filled computational workflow including one or more filled workflow stages. ¶ 0064: The configured composition is transformed into a service, with generated service code producing the composition with models as code.
DVORAK discloses computing the filled computational workflow includes computing respective filled values of the one or more fillable fields. ¶ 0049: Generation fills component template code or configured services through JSON files, environment variables, or API calls.
DVORAK discloses at least one of the filled values included in the filled computational workflow is executable code. ¶ 0049: Filled templates produce service code for composition models, including Python, C++, or scripts. ¶ 0064: The generated composition is compiled, deployed, and run.
DVORAK discloses execute the filled computational workflow. ¶ 0064: A configured composition can be “transformed into a service, deployed, and run.”
DVORAK does not expressly disclose the plurality of workflow stages including a plurality of unfilled workflow stages; computing respective filled values of the one or more fillable fields at least in part at a generative language model; and at least one of the filled values included in the filled computational workflow is a prompt of a machine learning model called in the corresponding filled workflow stage.
FRIDDLE discloses receive a computational workflow specification including: a plurality of workflow stages, the plurality of workflow stages including a plurality of unfilled workflow stages that each include: a respective workflow stage objective, a respective workflow stage exit criterion; and one or more fillable fields. ¶ 0271: The generation prompt supplies the process statement, process plan, and a commented JSON structure, and requests components of each individual step. ¶ 0151: The process data object contains a prompt and configuration component and a code component. ¶ 0164: There may be any number of logical blocks, and “all will have these components;” some steps need not be fully described. ¶ 0173: Logical blocks may contain pseudo instruction and subsequently be populated with relevant database information.
FRIDDLE discloses computing the filled computational workflow includes computing respective filled values of the one or more fillable fields at least in part at a generative language model. ¶ 0271: The LLM produces the JSON structure for individual steps. Deterministic logic then constructs the steps and assembles generated prompt components into a prompt.
FRIDDLE discloses at least one of the filled values included in the filled computational workflow is a prompt of a machine learning model called in the corresponding filled workflow stage. ¶ 0271: The plan is translated into “individual prompts and python functions needed at each step of the process.” ¶ 0161: Prompt 503 of logical block 515 “is the prompt that will get sent to the configured model.” ¶ 0272: The generated individual prompts from prompt file 514, and the corresponding Python functions are then constructed.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to modify the template based service generation of DVORAK to incorporate the generation of individual stage prompts and code from a process plan taught by FRIDDLE. One of ordinary skill in the art would be motivated to integrate that generation into DVORAK, with a reasonable expectation of success, in order to reduce token cost, generation time, and construction errors by generated structured components of each step and assembling them deterministically, as taught by FRIDDLE at ¶ 0271.
Claims 6 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over DVORAK, in view of FRIDDLE, further in view of Lin et al., U.S. PG-Publication No. 2024/0427566 A1 (hereinafter LIN).
Claim 6
LIN discloses during computation of the filled value of at least one of the unfilled workflow stages, the one or more processing devices are further configured to: perform retrieval-augmented generation (RAG) based at least in part on the workflow stage objective and/or the workflow stage exit criterion of the unfilled workflow stage to obtain a RAG query result. ¶ 0046: The expressly identified RAG module handles requests by “processing tasks and querying the code index and configuration module to augment LLM prompts.” Retrieved information becomes additional context for the large language model. ¶ 0051: The prompt module generates prompts for code generation and may “take the general task specified in project information as a base prompt and utilize RAG” to incorporate intervention or additional project information. ¶ 0065: The task statement described the programming task and its “high-level objectives,” such as resolving a resource leak or adding a feature.
LIN discloses compute the filled value based at least in part on the RAG query result. ¶ 0083: The service generates source code by providing a prompt to the model using project configuration and intervention information, with the prompt formed by the previously described prompt module. ¶ 0077: Source code produced by providing the prompt to an LLM” can be written to the repository and tested in a development environment.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to modify the stage implementation generation of DVORAK-FRIDDLE to incorporate the task based RAG code generation taught by LIN, using the selected stage objective to guide retrieval and the retrieved information to generates its filled code value. One of ordinary skill in the art would be motivated to integrate that RAG code generation into DVORAK-FRIDDLE, with a reasonable expectation of success, in order to reduce hallucinations in generated stage code by supplementing the model prompt with retrieved code and project information, as taught by LIN at ¶ 0046.
Claim 16
Claim 16 is rejected utilizing the rationale for claim 6; the claim is directed to a method performed by the system.
Claims 8 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over DVORAK, in view of FRIDDLE, further in view of Nicolae et al., U.S. PG-Publication No. 2025/0363371 A1 (hereinafter NICOLAE).
Claim 8
NICOLAE discloses wherein the directed graph structure includes a fan-in pattern, a fan-out pattern, and/or a map-reduce pattern. ¶ 0037: “pipeline stages are instead generalized as tasks” that can execute concurrently. A task receives input from multiple upstream tasks and sends output to multiple downstream tasks: “a task may have fan-in of inputs and fanout of outputs.” ¶ 0038: A task can send the same output to two downstream tasks or different outputs to respective downstream tasks. Each task may be a distinct graph vertex, with a directed edge carrying an upstream task’s output to a downstream task’s input. ¶ 0039: The workflow graph contains directed dataflow and synchronization edges. “A vertex is a representation of a task.” Synchronization edges prevent downstream execution until an upstream task finishes.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to modify the workflow graph orchestration of DVORAK-FRIDDLE to incorporate the task vertices with converging inputs and branching outputs taught by NICOLAE. One of ordinary skill in the art would be motivated to integrate those task connections into DVORAK-FRIDDLE, with a reasonable expectation of success, in order to reduce workflow latency and avoid underusing computing resources by permitting concurrent task execution instead of imposing a strict sequential ordering on all stages, as taught by NICOLAE at ¶ 0037.
Claim 18
Claim 18 is rejected utilizing the rationale for claim 8; the claim is directed to a method performed by the system.
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over DVORAK, in view of FRIDDLE, further in view of Sathianathan et al., U.S. PG-Publication No. 2025/0165775 A1 (hereinafter SATHIANATHAN).
Claim 9
SATHIANATHAN discloses wherein, during execution of the filled computational workflow, the one or more processing devices are configured to dynamically select an input format and/or an output format of at least one workflow stage of the plurality of workflow stages. ¶ 0158: The resulting workflow receives a browser request identifying the device as mobile or PC. “Responsive to the application input 750,” it generates the webpage in a “corresponding mobile or PC webpage format.” ¶ 0151: Conditional inputs define scenarios in which the application adapts its outputs .Selection between mobile and PC presentation depends on “the type of user device used.” ¶ 0173: Generated functionality modules are arranged into an “executable workflow” that generates outputs responsive to received inputs. ¶ 0111: A disclosed modular workflow ultimately passes results to “a functionality module that displays the results to the user,” for example by generating HTML for a browser.
It would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to modify the output presentation portion of DVORAK-FRIDDLE to incorporate the device responsive output format selection taught by SATHANATHAN. One of ordinary skill in the art would be motivated to integrate that selection into DVORAK-FRIDDLE, with a reasonable expectation of success, in order to avoid maintaining separate mobile and PC workflows by selecting the appropriate presentation format within one generated workflow, based on SATHIANATHAN at ¶¶ 0151, 0155, and 0158.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. See KEHRES, U.S. PG-Publication No. 2025/0272062 A1, teaching using LLMs to complete initially unconfigured workflow steps from user specified objectives, retrieve information for generating step conditions, and assemble the resulting steps into workflow code (¶¶ 0039, 0079, 0093, 0109, 0125).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to FRANK D MILLS whose telephone number is (571)270-3194. The examiner can normally be reached M-F 9-5:30 CT.
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, KEVIN YOUNG can be reached at (571)270-3180. 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.
/FRANK D MILLS/Primary Examiner, Art Unit 2194 October 15, 2026