System and method of maintaining contextual relationships and operational compatability in codebase functionality employed across multiple platforms
The inventive system of FIG. 1 and its underlying functionality make use of AI agents to support adaptation of data processing functions across different but inter-related operating platforms to maintain inter-operability of functions. AI agents determine, using a multi-dimensional knowledge graph and dependent sub-graphs, contextual relationships for information chunks associated with nodes and edges and classify these contextual relationships by weighting inferred for each mode by related sub-graphs. AI-agent interrogation of the sub-graphs identifies the extent and impact by which coding evolution in one platform adversely affects existing operational functionality in the different operating platform(s), or in an application in a data migration use case. The use of object-orientated codebases provides for self-learning within the resultant multi-dimensional knowledge graph, thereby permitting inquisitorial assessment of impact as well as appropriate code generation under control of system intelligence and, particularly, an AI-agent swarm. Code generation is independent of any front end LLM, with the generated code propagated automatically through the system to preserve inter-operability of functions instantiated in the inter-related operating platforms.
1 . A software development system including:
a plurality of trained AI-agents each having assigned functional tasks, said AI agents together forming an agent swarm;
a task manager configured to oversee interactions within the agent swarm;
a knowledge graph database responsive to the AI agents, the knowledge graph database containing an updateable multi-dimensional knowledge graph in which each node contains a data chunk and wherein each node and at least some edges between nodes are classified by at least one dependent cascaded sub-graph having an assigned attribute for a weighted significance of the data chunk, determined by the agent swarm, with respect to an identified operational impact of the data chunk in a plurality of code modules executed within at least one operational processing platform, wherein each assigned attribute is:
generated by the agent swarm by having AI agents interact to apply multi-hop reasoning that reflects a determination of how a first data chunk instantiated in a codebase has previously affected functionality associated with use of the first data chunk; and
updateable, as determined by the agent swarm, with time; and
processing intelligence, responsive to the multi-dimensional knowledge graph and associated with the agent swarm, wherein the processing intelligence is configured to manage a code writing function configured to take into account, in the generation of code, the assigned attributes for nodes or edges stored within cascaded sub-graphs of the multi-dimensional knowledge graph,
wherein the processing intelligence is arranged to, one of:
(i) mitigate one or more effects of instantiated or generated code, making direct use of data chunks or based on evolution of the data chunks, determined by the agent swarm and historically recorded by the processing intelligence to have caused adversely compromised processing of data or data integrity within an operational processing platform accessible to the processing intelligence; and
(ii) prevent generation of code making direct use of data chunks determined by the agent swarm historically recorded by the processing intelligence to have caused functional disruption within an operational processing platform accessible to the processing intelligence.
2 . The software development system of claim 1 , wherein the code writing function is independent of any front-end large language model “LLM,” and wherein
the processing intelligence is arranged automatically to propagate generated code, to inter-operationally related operating platforms, that preserves inter-operability of functions instantiated in those inter-related operating platforms.
3 . The software development system of claim 2 , wherein the agent swarm, in applying multi-hop reasoning, is configured to determine how data chunks, instantiated in a codebase, have previously affected similar functional tasks, instantiated in a plurality of operational processing platforms, to that of the first data chunk associated with use of the first data chunk in at least one operational processing platform.
4 . The software development system of claim 1 , wherein the agent swarm, in applying multi-hop reasoning, is configured to determine how data chunks, instantiated in a codebase, have previously affected similar functional tasks, instantiated in a plurality of operational processing platforms, to that of the first data chunk associated with use of the first data chunk in at least one operational processing platform.
5 . The software development system of claim 1 , wherein the processing intelligence is arranged to feed back any mitigation strategy into the agent swarm to cause update of the multi-dimensional knowledge graph, thereby to enhance its history and support predictive capabilities of the AI agents.
6 . The software development system of claim 1 , wherein the attributes are contextual relationships.
7 . The software development system of claim 1 , wherein the system uses object-orientated codebases to support self-learning within the multi-dimensional knowledge graph, and the system further includes:
a user interface arranged to input a query and to receive a response thereto.
8 . The software development system of claim 7 , wherein the response is sent as a message over a network.
9 . The software development system according to claim 1 , wherein the processing intelligence is a distributed intelligence.
10 . The software development system according to claim 1 , wherein the processing intelligence is configured to break down and distribute one or more tasks to the AI agents.
11 . A software development system including:
a plurality of trained AI-agents each having assigned functional tasks, said AI agents together forming an agent swarm;
a task manager configured to oversee interactions within the agent swarm;
a knowledge graph database responsive to the AI agents, the knowledge graph database containing an updateable multi-dimensional knowledge graph in which each node contains a data chunk and wherein each node and at least some edges between nodes are classified by at least one dependent cascaded sub-graph having an assigned attribute for a weighted significance of the data chunk, determined by the agent swarm, with respect to an identified operational impact of the data chunk in a plurality of code modules executed within at least one operational processing platform, wherein each assigned attribute is:
generated by the agent swarm by having AI agents interact to apply multi-hop reasoning that reflects a determination of how a first data chunk instantiated in a codebase has previously affected functionality associated with use of the first data chunk; and
updateable, as determined by the agent swarm, with time; and
processing intelligence, responsive to the multi-dimensional knowledge graph and associated with the agent swarm, wherein the processing intelligence is configured to manage a code writing function configured to take into account, in the generation of code, the assigned attributes for nodes or edges stored within cascaded sub-graphs of the multi-dimensional knowledge graph,
wherein a weight, w, is determined by the agent swarm and assigned, by the agent swarm, to an attribute in a sub-graph for each node according to a function
w (node)=α N _dep+βΣedge_crit+γchange_freq
where:
w(node) for a sub-graph for a specific node;
N_dep is the number of direct dependencies;
edge_crit is a criticality score learned from post-mortem incident data reporting efficacy and positive/negative effects of a change at the node or change in the chunk;
change_freq captures how often a chunk has been changed; and
α, β and β are integers.
12 . A software development system including:
a plurality of trained AI-agents each having assigned functional tasks, said AI agents together forming an agent swarm;
a task manager configured to oversee interactions within the agent swarm;
a knowledge graph database responsive to the AI agents, the knowledge graph database containing an updateable multi-dimensional knowledge graph in which each node contains a data chunk and wherein each node and at least some edges between nodes are classified by at least one dependent cascaded sub-graph having an assigned attribute for a weighted significance of the data chunk, determined by the agent swarm, with respect to an identified operational impact of the data chunk in a plurality of code modules executed within at least one operational processing platform, wherein each assigned attribute is:
generated by the agent swarm by having AI agents interact to apply multi-hop reasoning that reflects a determination of how a first data chunk instantiated in a codebase has previously affected functionality associated with use of the first data chunk; and
updateable, as determined by the agent swarm, with time; and
processing intelligence, responsive to the multi-dimensional knowledge graph and associated with the agent swarm, wherein the processing intelligence is configured to manage a code writing function configured to take into account, in the generation of code, the assigned attributes for nodes or edges stored within cascaded sub-graphs of the multi-dimensional knowledge graph;
wherein the code writing function is independent of any front-end large language model “LLM,”
wherein the processing intelligence is arranged automatically to propagate generated code, to inter-operationally related operating platforms, that preserves inter-operability of functions instantiated in those inter-related operating platforms,
wherein the agent swarm, in applying multi-hop reasoning, is configured to determine how data chunks, instantiated in a codebase, have previously affected similar functional tasks, instantiated in a plurality of operational processing platforms, to that of the first data chunk associated with use of the first data chunk in at least one operational processing platform, and
wherein a weight, w, is determined by the agent swarm and assigned, by the agent swarm, to an attribute in a sub-graph for each node according to a function
w (node)=α N _dep+βΣedge_crit+γchange_freq
where:
w(node) for a sub-graph for a specific node;
N_dep is the number of direct dependencies;
edge_crit is a criticality score learned from post-mortem incident data reporting efficacy and positive/negative effects of a change at the node or change in the chunk;
change_freq captures how often a chunk has been changed; and
α, β and γ are integers.
13 . A software development system including:
a plurality of trained AI-agents each having assigned functional tasks, said AI agents together forming an agent swarm;
a task manager configured to oversee interactions within the agent swarm;
a knowledge graph database responsive to the AI agents, the knowledge graph database containing an updateable multi-dimensional knowledge graph in which each node contains a data chunk and wherein each node and at least some edges between nodes are classified by at least one dependent cascaded sub-graph having an assigned attribute for a weighted significance of the data chunk, determined by the agent swarm, with respect to an identified operational impact of the data chunk in a plurality of code modules executed within at least one operational processing platform, wherein each assigned attribute is:
generated by the agent swarm by having AI agents interact to apply multi-hop reasoning that reflects a determination of how a first data chunk instantiated in a codebase has previously affected functionality associated with use of the first data chunk; and
updateable, as determined by the agent swarm, with time; and
processing intelligence, responsive to the multi-dimensional knowledge graph and associated with the agent swarm, wherein the processing intelligence is configured to manage a code writing function configured to take into account, in the generation of code, the assigned attributes for nodes or edges stored within cascaded sub-graphs of the multi-dimensional knowledge graph;
wherein the code writing function is independent of any front-end large language model “LLM,”
wherein the processing intelligence is arranged automatically to propagate generated code, to inter-operationally related operating platforms, that preserves inter-operability of functions instantiated in those inter-related operating platforms,
wherein the agent swarm, in applying multi-hop reasoning, is configured to determine how data chunks, instantiated in a codebase, have previously affected similar functional tasks, instantiated in a plurality of operational processing platforms, to that of the first data chunk associated with use of the first data chunk in at least one operational processing platform, and
wherein the processing intelligence is arranged to, one of:
(i) mitigate one or more effects of instantiated or generated code, making direct use of data chunks or based on evolution of the data chunks, determined by the agent swarm ( 102 ) and historically recorded by the processing intelligence to have caused adversely compromised processing of data or data integrity within an operational processing platform accessible to the processing intelligence; and
(ii) prevent generation of code making direct use of data chunks determined by the agent swarm ( 102 ) historically recorded by the processing intelligence to have caused functional disruption within an operational processing platform accessible to the processing intelligence.
14 . A data migration system comprising:
the software development system of claim 1 ,
a user interface arranged to allow oversight of the software development system; and
wherein the system is arranged to create a model based on a transformer architecture trained on a corpus of regular text to predict a relationship between source and target data on at least one of a field-to-field mapping and value mapping within fields.
15 . The data migration system of claim 14 , wherein the code writing function of the software development system is independent of any front-end large language model “LLM,” and wherein
the processing intelligence is arranged automatically to propagate generated code, to inter-operationally related operating platforms, that preserves inter-operability of functions instantiated in those inter-related operating platforms.
16 . The data migration system of claim 14 , wherein the agent swarm of the software development system, in applying multi-hop reasoning, is configured to determine how data chunks, instantiated in a codebase, have previously affected similar functional tasks, instantiated in a plurality of operational processing platforms, to that of the first data chunk associated with use of the first data chunk in at least one operational processing platform.
17 . The data migration system of claim 14 , wherein a weight, w, is determined by the agent swarm of the software development system and assigned, by the agent swarm, to an attribute in a sub-graph for each node according to a function
w (node)=α N _dep+βΣedge_crit+γchange_freq
where:
w(node) for a sub-graph for a specific node;
N_dep is the number of direct dependencies;
edge_crit is a criticality score learned from post-mortem incident data reporting efficacy and positive/negative effects of a change at the node or change in the chunk;
change_freq captures how often a chunk has been changed; and
α, β and γ are integers.
18 . A method of developing software code for operation in an operational software platform, the method comprising:
instantiating a similarity search by traversing a multi-dimensional knowledge graph containing cascaded dependent sub-graphs in which each node at a highest level of the multi-dimensional knowledge graph contains a data chunk and wherein at least each node is classified by at least one cascaded sub-graph depending from the node and present within a lower contextual level of the multi-dimensional knowledge graph; and
assigning, to sub-graphs, attributes for contextually weighted significance of a respective data chunk, determined by interacting AI agents arranged in an agent swarm and wherein each AI agent each has an assigned functional task, with respect to an identified operational impact of the respective data chunk in a code module executed within an operational software platform,
the method further comprising at least one of:
mitigating, in response to assessment by system intelligence of historically monitored effects related to use of a first data chunk in a code module, adverse functional effects arising from use in code generated using the first data chunk, wherein mitigating is one of:
i) suppression of the use of the first data chunk in altered coding; and
ii) generation and communication of a warning of how the data chunk could affect functionality of at least one code module in at least one inter-related operating platform.
19 . The method of claim 18 , further comprising:
using a task manager to oversee and manage process interactions between AI agents in the agent swarm; and
applying multi-hop reasoning within the AI swarm to determine how data chunks, instantiated in a codebase, have previously affected functionality in at least one software platform.
20 . The method of claim 19 , further comprising:
updating the sub-graphs to reflect predicted or actual changes in operational functionality, determined by the agent swarm, in response to modification of a first data chunk of a node and instantiation of the first chunk in a plurality of code modules in a plurality of inter-related platforms.
21 . The method of claim 20 , wherein the agent swarm, in applying multi-hop reasoning, determines how data chunks have previously affected similar functional tasks, instantiated in a plurality of operational software platforms.
22 . The method of claim 18 , further comprising:
updating the sub-graphs to reflect predicted or actual changes in operational functionality, determined by the agent swarm, in response to modification of a first data chunk of a node and instantiation of the first chunk in a plurality of code modules in a plurality of inter-related platforms.
23 . The method of claim 18 , further comprising:
executing a code writing function, using the agent swarm, that takes into account, in the generation of code, the assigned attributes associated with nodes or edges stored within cascaded sub-graphs of the multi-dimensional knowledge graph.
24 . The method of claim 23 , wherein the code writing function is independent of any front-end large language model “LLM”.
25 . The method of claim 23 , wherein generated code preserves inter-operability of functions to be instantiated into inter-related operating software platforms.
26 . The method of claim 18 , wherein the agent swarm, in applying multi-hop reasoning, determines how data chunks have previously affected similar functional tasks, instantiated in a plurality of operational software platforms.
27 . The method of claim 18 , further comprising:
feeding back any mitigation strategy into the agent swarm to update the multi-dimensional knowledge graph, thereby enhancing its history and supporting predictive capabilities of the AI agents.
28 . A method of developing software code for operation in an operational software platform, the method comprising:
instantiating a similarity search by traversing a multi-dimensional knowledge graph containing cascaded dependent sub-graphs in which each node at a highest level of the multi-dimensional knowledge graph contains a data chunk and wherein at least each node is classified by at least one cascaded sub-graph depending from the node and present within a lower contextual level of the multi-dimensional knowledge graph; and
assigning, to sub-graphs, attributes for contextually weighted significance of a respective data chunk, determined by interacting AI agents arranged in an agent swarm and wherein each AI agent each has an assigned functional task, with respect to an identified operational impact of the respective data chunk in a code module executed within an operational software platform,
wherein the agent swarm determines a weight, w, to be assigned a sub-graph for each node, and the weight is determined according to a function
w (node)=α N _dep+βΣedge_crit+γchange_freq
where:
w(node) for a sub-graph for a specific node;
N_dep is the number of direct dependencies;
edge_crit is a criticality score learned from post-mortem incident data reporting efficacy and positive/negative effects of a change at the node or change in the chunk;
change_freq captures how often a chunk has been changed; and
α, β and γ are integers.
29 . A method of developing software code for operation in an operational software platform, the method comprising:
instantiating a similarity search by traversing a multi-dimensional knowledge graph containing cascaded dependent sub-graphs in which each node at a highest level of the multi-dimensional knowledge graph contains a data chunk and wherein at least each node is classified by at least one cascaded sub-graph depending from the node and present within a lower contextual level of the multi-dimensional knowledge graph; and
assigning, to sub-graphs, attributes for contextually weighted significance of a respective data chunk, determined by interacting AI agents arranged in an agent swarm and wherein each AI agent each has an assigned functional task, with respect to an identified operational impact of the respective data chunk in a code module executed within an operational software platform,
the method further comprising updating the sub-graphs to reflect predicted or actual changes in operational functionality, determined by the agent swarm, in response to modification of a first data chunk of a node and instantiation of the first chunk in a plurality of code modules in a plurality of inter-related platforms, and
the method further comprising at least one of:
mitigating, in response to assessment by system intelligence of historically monitored effects related to use of a first data chunk in a code module, adverse functional effects arising from use in code generated using the first data chunk, wherein mitigating is one of:
i) suppression of the use of the first data chunk in altered coding; and
ii) generation and communication of a warning of how the data chunk could affect functionality of at least one code module in at least one inter-related operating platform.
30 . A method of developing software code for operation in an operational software platform, the method comprising:
instantiating a similarity search by traversing a multi-dimensional knowledge graph containing cascaded dependent sub-graphs in which each node at a highest level of the multi-dimensional knowledge graph contains a data chunk and wherein at least each node is classified by at least one cascaded sub-graph depending from the node and present within a lower contextual level of the multi-dimensional knowledge graph; and
assigning, to sub-graphs, attributes for contextually weighted significance of a respective data chunk, determined by interacting AI agents arranged in an agent swarm and wherein each AI agent each has an assigned functional task, with respect to an identified operational impact of the respective data chunk in a code module executed within an operational software platform,
the method further comprising:
using a task manager to oversee and manage process interactions between AI agents in the agent swarm;
applying multi-hop reasoning within the AI swarm to determine how data chunks, instantiated in a codebase, have previously affected functionality in at least one software platform; and
updating the sub-graphs to reflect predicted or actual changes in operational functionality, determined by the agent swarm, in response to modification of a first data chunk of a node and instantiation of the first chunk in a plurality of code modules in a plurality of inter-related platforms,
wherein the agent swarm, in applying multi-hop reasoning, determines how data chunks have previously affected similar functional tasks, instantiated in a plurality of operational software platforms, and
wherein the agent swarm determines a weight, w, to be assigned a sub-graph for each node, and the weight is determined according to a function
w (node)=α N _dep+βΣedge_crit+γchange_freq
where:
w(node) for a sub-graph for a specific node;
N_dep is the number of direct dependencies;
edge_crit is a criticality score learned from post-mortem incident data reporting efficacy and positive/negative effects of the change at the node or change in the chunk;
change_freq captures how often a chunk has been changed; and
α, β and γ are integers.