Method and system for multi-level artificial intelligence supercomputer design

1. A method of answering queries using one or more families of large language models (h-LLMs) by a computer comprising a processor, a non-transitory storage medium, and software on the storage medium, the method comprising:

receiving a plurality of user prompts via an application programming interface (API) at a query layer from a user interface;

determining an analysis mode of the plurality of user prompts to be a batch mode;

transmitting a batch query prompt comprising the plurality of user prompts from the query layer to a batch layer responsive to determining the analysis mode being a batch mode, the batch layer comprising at least one h-LLM;

receiving a batch response from the batch layer responsive to the batch query prompt; and

transmitting the batch response to the user interface via the API.

2. The method of claim 1 wherein the batch query prompt comprises an indication of one or more of the quantity of user prompts comprised by the batch query prompt or a time interval for processing the batch query prompt.

3. The method of claim 2 wherein the plurality of user prompts is received in a single batch of user prompts from the user interface.

4. The method of claim 1 wherein:

the plurality of user prompts is received at an input broker comprised by the query layer;

the batch response is received by an output broker comprised by the query layer; and

the batch response is transmitted by the output broker.

5. The method of claim 4 wherein the batch response is transmitted as one of a plurality of batch responses collected by the output broker and comprised by a single batch response transmission.

6. The method of claim 1 wherein the batch query prompt is transmitted to the batch layer as part of a data batch comprising a plurality of query prompts to the batch layer.

7. The method of claim 1 wherein the at least one h-LLM comprised by the batch layer is in a cloud container environment.

8. The method of claim 7 wherein the cloud container environment containing the at least one h-LLM comprises a microservice.

9. The method of claim 7 wherein the cloud container environment is implemented using a Kubernetes environment and a service mesh, providing at least one containerized h-LLM microservice.

10. The method of claim 1 wherein the at least one h-LLM comprised by the batch layer is fine-tuned from a base h-LLM by data batches received from an input user data stream.

11. A method of answering queries using one or more families of large language models (h-LLMs) by a computer comprising a processor, a non-transitory storage medium, and software on the storage medium, the method comprising:

receiving a plurality of user prompts via an application programming interface (API) at a query layer from a user interface;

determining an analysis mode of the plurality of user prompts to be one of a batch mode or a real-time mode;

transmitting a query prompt from the query layer to at least one of a batch layer comprising at least one h-LLM and a real-time layer comprising at least one h-LLM responsive to the plurality of user prompts;

receiving at least one of a batch response from the batch layer and a real-time response from the real-time layer; and

transmitting the at least one of the batch response and the real-time response to the user interface via the API.

12. The method of claim 11 wherein:

the plurality of user prompts is received at an input broker comprised by the query layer;

the at least one of the batch response and the real-time response is received by an output broker comprised by the query layer; and

the at least one of the batch response and the real-time response is transmitted by the output broker through use of a microservice executed on a graphics processing unit (GPU)-based container.

13. The method of claim 12 wherein the batch response is transmitted as one of a plurality of batch responses collected by the output broker and comprised by a single batch response transmission.

14. The method of claim 11 wherein the query prompt is transmitted to the batch layer as part of a data batch comprising a plurality of query prompts to the batch layer.

15. The method of claim 11 wherein the at least one h-LLM comprised by the batch layer or the at least one h-LLM comprised by the real-time layer is in a cloud container environment.

16. The method of claim 15 wherein the cloud container environment containing the at least one h-LLM comprised by the batch layer or the at least one h-LLM comprised by the real-time layer comprises a microservice.

17. The method of claim 15 wherein the cloud container environment:

is implemented using a Kubernetes environment and a service mesh;

provides at least one containerized h-LLM microservice; and

is hosted on at least one GPU.

18. The method of claim 11 wherein the at least one h-LLM comprised by the batch layer is fine-tuned from a base h-LLM by data batches received from an input data stream.

19. The method of claim 11 wherein the at least one h-LLM comprised by the real-time layer is fine-tuned from a base h-LLM by incremental updates received in real-time from an input data stream.

20. A system for answering queries using one or more families of large language models (h-LLMs) comprising:

a processor configured to:

operate a query layer configured to:

receive a plurality of user prompts via an application programming interface (API) from a user interface;

determine an analysis mode of the plurality of user prompts to be a batch mode;

transmit a batch query prompt from the query layer to a batch layer responsive to the plurality of user prompts, the batch layer comprising at least one h-LLM;

receive a batch response from the batch layer responsive to the batch query prompt; and

transmit the batch response to the user interface via the API; and

operate the batch layer comprising at least one h-LLM, the batch layer being configured to process the batch query prompt, thereby generating the batch response;

a non-transitory computer-readable storage medium having stored thereon software that is executable by the processor; and

a communication device operable to enable communication between the processor and the user interface.