AI-powered virtual tag coverage expansion

1 . A computer implemented method of automatically generating rules for expanding coverage of existing virtual tags and/or for generation of new virtual tags of a cloud computing environment, comprising:

managing a database of resource records, wherein each resource record includes a resource value indicating utilization of a resource of a plurality of resources of the cloud computing environment, wherein the database includes a plurality of untagged resource records and a plurality of tagged resource records;

extracting the plurality of untagged resource records from the database;

generating a key-value distribution associating aggregated resource values with tag key/tag value pairs extracted from each of the plurality of untagged resource records;

generating a structured prompt including: existing rules for generating virtual tags, the key-value distribution, and instructions for identifying tag keys of the plurality of untagged resource records that are semantically related to a concept represented by virtual tags generated based on the existing rules, and for computing at least one new rule according to the identified tag keys, the at least one new rule is for generating at least one new virtual tag and/or for expanding coverage of at least one existing virtual tag;

feeding the structured prompt into a large language model (LLM);

receiving the at least one new rule from the LLM; and

adding the at least one new rule to a rule repository associated with the database for application to the plurality of untagged resource records and to new untagged resource records.

2 . The computer implemented method of claim 1 , further comprising:

receiving a new resource record generated for the cloud computing environment, wherein the new resource record is untagged;

applying the at least one new rule to the new resource record for generating the new virtual tag and/or expanding coverage of the at least one existing virtual tag;

storing an association between the new resource record and the new virtual tag and/or expanded existing virtual tag in the database.

3 . The computer implemented method of claim 2 , wherein the new resource record is automatically generated in response to detected utilization of a resource of the plurality of resources.

4 . The computer implemented method of claim 1 , wherein the at least one new rule comprises a plurality of candidate new rules, and further comprising:

for each respective candidate rule of the plurality of candidate rules, simulating application of the respective candidate rule to the plurality of untagged resource records to simulate generation of virtual tags to obtain a simulated metric;

ranking the plurality of new candidate rules according to corresponding simulated metrics; and

selecting a highest ranked candidate rule,

wherein the at least one new rule added to a rule repository comprises the highest ranked candidate rule.

5 . The computer implemented method of claim 4 , further comprising:

computing a baseline total of a plurality of resource values for the plurality of untagged resource records;

for each respective candidate rule of the plurality of candidate rules:

the simulated metric comprises a simulated total of the plurality of resource values for a subset of the plurality of untagged records for which virtual tags were computed by the simulated application of the respective candidate rule;

computing an absolute improvement provided by the simulated total for the respective candidate rule over the baseline total,

wherein ranking comprises ranking the plurality of candidate rules according to corresponding absolute improvement,

wherein the highest ranked candidate rule comprises the candidate rule with highest absolute improvement,

wherein the at least one new rule added to the rule repository comprises the candidate rule with highest absolute improvement.

6 . The computer implemented method of claim 4 , further comprising:

computing a baseline total of a plurality of resource values for the plurality of untagged resource records;

for each respective candidate rule of the plurality of candidate rules:

the simulated metric comprises a simulated total of the plurality of resource values for a subset of the plurality of untagged records for which virtual tags were computed by the simulated application of the respective candidate rule;

computing a relative improvement provided by the simulated total for the respective candidate rule with respect to the baseline total,

wherein ranking comprises ranking the plurality of candidate rules according to corresponding relative improvement,

wherein the highest ranked candidate rule comprises the candidate rule with highest relative improvement,

wherein the at least one new rule added to the rule repository comprises the candidate rule with highest relative improvement.

7 . The computer implemented method of claim 1 , wherein the database includes tagged resource records tagged with virtual tags generated according to a plurality of existing rules hosted in a rule repository.

8 . The computer implemented method of claim 1 , wherein extracting the plurality of untagged resource records from the database comprises generating a query for searching the database for untagged resource records, and executing the query to extract the plurality of untagged resource records.

9 . The computer implemented method of claim 1 , wherein the database comprises a cost database, wherein each record comprises a cost record, and the resource value comprises a cost of using the resource.

10 . The computer implemented method of claim 1 , further comprising filtering tag keys to generate filtered tag keys that exclude cloud-provider system tag keys that are unrelated to a target concept, wherein the key-value distribution is generated from the filtered tag keys.

11 . The computer implemented method of claim 1 , wherein the structured prompt further includes instructions for analyzing the existing rules to extract a plurality of target concepts being tagged and for generating the at least one new rule to generate the at least one virtual tag based on the plurality of target concepts.

12 . The computer implemented method of claim 1 , wherein the structured prompt includes instructions for analyzing semantic relationships between: patterns of existing rules for generating virtual tags, available tag keys in the untagged resource records, and value patterns that suggest conceptual alignment.

13 . The computer implemented method of claim 1 , wherein the LLM identifies at least one candidate tag key likely to be associated with values of the candidate tag key representing same or corresponding target concepts, wherein the LLM generates the at least one new rule based on the at least one candidate tag key.

14 . The computer implemented method of claim 1 , wherein the LLM generates a rule specification including each new rule and at least one of: filter conditions, target allocation mapping, and reasoning explaining why a candidate tag key associated with the new rule is semantically relevant.

15 . The computer implemented method of claim 14 , wherein the rule specification is generated for complying with a format of rules stored in a rule repository applied to new resource records.

16 . The computer implemented method of claim 1 , wherein the LLM normalizes the tag values to comply with an existing taxonomy of the virtual tag.

17 . The computer implemented method of claim 1 , wherein for tag keys that do not exactly match existing allocations of existing virtual tags, the LLM performs a semantic similarity analysis to generate mappings between observed tag values and existing allocation values, wherein the mappings are designed to preserve an existing taxonomy implemented by existing virtual tags while expanding coverage for including at least one new virtual tag and/or expanding coverage of the at least one existing virtual tag generated by the at least one new rule.

18 . A system for automatically generating rules for expanding coverage of existing virtual tags and/or for generation of new virtual tags of a cloud computing environment, comprising:

at least one processor executing a code for:

managing a database of resource records, wherein each resource record includes a resource value indicating utilization of a resource of a plurality of resources of the cloud computing environment, wherein the database includes a plurality of untagged resource records and a plurality of tagged resource records;

extracting the plurality of untagged resource records from the database;

generating a key-value distribution associating aggregated resource values with tag key/tag value pairs extracted from each of the plurality of untagged resource records;

generating a structured prompt including: existing rules for generating virtual tags, the key-value distribution, and instructions for identifying tag keys of the plurality of untagged resource records that are semantically related to a concept represented by virtual tags generated based on the existing rules, and for computing at least one new rule according to the identified tag keys, the at least one new rule is for generating at least one new virtual tag and/or for expanding coverage of at least one existing virtual tag;

feeding the structured prompt into a large language model (LLM);

receiving the at least one new rule from the LLM; and

adding the at least one new rule to a rule repository associated with the database for application to the plurality of untagged resource records and to new untagged resource records.

19 . A non-transitory medium storing program instructions for expanding coverage of existing virtual tags and/or for generation of new virtual tags of a cloud computing environment, which when executed by at least one processor, cause the at least one processor to:

manage a database of resource records, wherein each resource record includes a resource value indicating utilization of a resource of a plurality of resources of the cloud computing environment, wherein the database includes a plurality of untagged resource records and a plurality of tagged resource records;

extract the plurality of untagged resource records from the database;

generate a key-value distribution associating aggregated resource values with tag key/tag value pairs extracted from each of the plurality of untagged resource records;

generate a structured prompt including: existing rules for generating virtual tags, the key-value distribution, and instructions for identifying tag keys of the plurality of untagged resource records that are semantically related to a concept represented by virtual tags generated based on the existing rules, and for computing at least one new rule according to the identified tag keys, the at least one new rule is for generating at least one new virtual tag and/or for expanding coverage of at least one existing virtual tag;

feed the structured prompt into a large language model (LLM);

receive the at least one new rule from the LLM; and

add the at least one new rule to a rule repository associated with the database for application to the plurality of untagged resource records and to new untagged resource records.