COPILOT IMPLEMENTATION: TESTING CONFORMANCE WITH KNOWLEDGE DOMAIN
Apparatus and methods are disclosed for implementing a copilot as a network of microservices including specialized large language models (LLMs) or other trained machine learning (ML) tools. The microservice network architecture supports flexible, customizable, or dynamically determinable dataflow. Compared to much larger competing LLMs, comparable or superior performance is achieved for certain tasks, while significantly reducing computation time and hardware requirements, even to a single compute node with a single GPU. Examples incorporate a qualification microservice to test data, destined for a downstream microservice, for conformance with the copilot's competency. A knowledge graph of a target domain is built, visualized, and pruned. The data is tested for conformance with the pruned domain representation and non-conforming data is discarded. Variations and additional techniques are disclosed.
1 . A method of filtering data within a copilot, comprising:
building a knowledge graph of a target domain, the knowledge graph comprising a plurality of vertices;
pruning a visualization of the knowledge graph to obtain a pruned domain representation, wherein the pruning deletes one or more portions of the knowledge graph;
subsequently, at a qualification microservice within the copilot:
receiving, for testing, the data outputted from another microservice within the copilot; and
testing the data for conformance with the pruned domain representation; and
discarding non-conforming data.
2 . The method of claim 1 , wherein the knowledge graph is built based on vector representations of documents of the target domain.
3 . The method of claim 1 , wherein the pruned domain representation comprises vector representations derived from the pruned visualization of the knowledge graph.
4 . The method of claim 1 , wherein the conformance is determined based on:
the tested data lying within an envelope defined based on nodes of the pruned visualization of the knowledge graph; or
a distance measure between the tested data and a node of the knowledge graph being less than or equal to a predetermined threshold.
5 . The method of claim 1 , wherein the data is obtained directly or indirectly from a retrieval microservice and conforming data is forwarded toward a core microservice.
6 . The method of claim 1 , wherein the data is received directly or indirectly from a core microservice and conforming data is forwarded toward an evaluation microservice.
7 . The method of claim 1 , wherein the data is in one or more of: an audio mode, an image mode, a numerical mode, or a text mode.
8 . One or more computer-readable media storing instructions which, when executed by one or more hardware processors, cause the one or more hardware processors to perform first operations comprising:
building a knowledge graph of a target domain of a copilot; and
pruning a visualization of the knowledge graph to obtain a pruned domain representation, wherein the pruning deletes one or more portions of the knowledge graph;
wherein the pruned domain representation is configured to be used by a qualification microservice of the copilot to perform second operations comprising:
testing data received by the qualification microservice for conformance with the pruned domain representation; and
discarding non-conforming data.
9 . The one or more computer-readable media of claim 8 , wherein the knowledge graph is built based on vector representations of documents of the target domain.
10 . The one or more computer-readable media of claim 8 , wherein the pruned domain representation comprises vector representations, and the pruning comprises:
deriving the vector representations from the pruned visualization of the knowledge graph.
11 . The one or more computer-readable media of claim 8 , wherein the stored instructions, when executed by the one or more hardware processors, further cause the one or more hardware processors to perform the second operations.
12 . The one or more computer-readable media of claim 8 , wherein the pruning is performed interactively.
13 . A system comprising:
one or more hardware processors, with memory coupled thereto; and
one or more computer readable media storing instructions comprising a plurality of modules which, when executed by the one or more hardware processors, implement respective microservices, the microservices forming a weakly connected network of microservices configured as a copilot for one or more first client applications;
wherein each of the microservices is configured to:
receive input from (i) a respective first group comprising one or more others of the microservices or (ii) one or more second client applications; and
transmit output to (i) a second group comprising one or more of the microservices or (ii) one or more third client applications;
wherein a plurality of the microservices incorporate respective trained machine learning tools;
wherein the network of microservices comprises at least a retrieval microservice, a qualification microservice, and one or more core microservices; and
wherein the qualification microservice is configured to:
test data received by the qualification microservice for conformance with a pruned domain representation, the pruned domain representation being derived from a visualization of a knowledge graph of a target domain of the copilot; and
discard non-conforming data.
14 . The system of claim 13 , wherein the conformance is determined based on:
the tested data lying within an envelope defined based on nodes of a pruned visualization of the knowledge graph.
15 . The system of claim 13 , wherein the conformance is determined based on:
a distance measure between the tested data and a node of the knowledge graph being less than or equal to a predetermined threshold.
16 . The system of claim 13 , wherein the received data is based on output of the retrieval microservice, and the qualification microservice is further configured to:
direct conforming data toward the one or more core microservices.
17 . The system of claim 13 , wherein the network of microservices further comprises an evaluation microservice, the received data is based on output of the one or more core microservices, and the qualification microservice is further configured to:
direct conforming data toward the evaluation microservice.
18 . The system of claim 13 , wherein the network of microservices further comprises an intermodal microservice or a data producer from which the data is received by the qualification microservice, and the qualification microservice is further configured to:
direct conforming data toward the retrieval microservice.
19 . The system of claim 13 , wherein the data is in two or more of: an audio mode, an image mode, a numerical mode, or a text mode.
20 . The system of claim 13 , wherein the copilot further comprises a reinforcement learning subsystem, and the system is configured to:
forward the non-conforming data toward the reinforcement learning subsystem for inclusion in a training record for the reinforcement learning subsystem.