Automated text-to-optimization routing
A request standardization system receives an optimization request from a user and generates a standardized representation of that optimization request. An optimization router parses the standardized representation of the optimization request to determine what type of optimization is being requested. The optimization router identifies a cluster of optimizers that should be invoked to generate proposed optimizations and routes the standardized representation to the identified cluster of optimizers. The optimization cluster generates a set of the proposed optimizations that are output.
1 . A computer implemented method, comprising:
conducting a chat user experience with a generative artificial intelligence (AI) system to obtain user input information indicative of an optimization request;
extracting, by the generative AI system from the chat user experience, characteristics of the optimization request from the user input information, the characteristics comprising a goal type that indicates a type of optimization goal and a constraint type that indicates a type of constraint in the optimization request;
generating a standardized representation of the optimization request, based on the extracted characteristics of the optimization request, the standardized representation including type information indicative of a type of optimization requested by the optimization request;
routing the standardized representation of the optimization request to an optimizer based on the type information, wherein routing is performed by consulting mappings that map the optimization goal type and the constraint types in the standardized representation to a cluster of optimizers;
running an optimization algorithm at the optimizer to generate an optimization result; and
generating an output indicative of the optimization result.
2 . The computer implemented method of claim 1 wherein receiving user input information comprises:
conducting a natural-language conversation with a transformer-based large language model (LLM), wherein the LLM extracts, from the chat user experience, the goal type and the constraint type, and generates a JSON document conforming to a predefined schema, the JSON document including the extracted goal type and the constraint type, and wherein a trained neural-network classifier receives the JSON document as input and outputs a cluster identifier that determines which optimizer cluster receives the standardized optimization request.
3 . The computer implemented method of claim 2 wherein generating the standardized representation of the optimization request comprises:
configuring the characteristics of the optimization request according to a predefined schema to obtain the standardized representation of the optimization request, wherein the standardized representation comprises the JSON document including fields for the goal type and the constraint type, each field populated with a value selected from a predefined taxonomy of optimization categories; and
wherein routing the standardized representation to the optimizer is performed by consulting a mapping datastore that maps the goal type code in the JSON document to a corresponding optimizer.
4 . The computer implemented method of claim 3 wherein configuring the characteristics comprises:
generating, by the generative AI system, a vector embedding of the entire optimization request based on the extracted characteristics,
wherein routing the standardized representation to the optimizer comprises performing a similarity search in an embedding index to select the optimizer cluster whose prototype vector is closest to the request embedding.
5 . The computer implemented method of claim 3 wherein conducting a user experience with the generative AI system comprises:
conducting the user experience with the generative AI system to obtain, as the characteristics of the optimization request, a decision variable type that indicates a type of decision variable in the optimization request.
6 . The computer implemented method of claim 5 wherein routing the standardized representation comprises:
parsing the standardized representation to obtain the characteristics of the optimization request; and
routing the standardized representation of the optimization request based on the characteristics of the optimization request.
7 . The computer implemented method of claim 6 wherein routing the standardized representation of the optimization request based on the characteristics of the optimization request comprises:
receiving the characteristics of the optimization request as inputs to a classifier;
generating, as a classifier output, a cluster identifier identifying a cluster of optimizers based on the inputs to the classifier; and
routing the standardized representation based on the classifier output.
8 . The computer implemented method of claim 6 wherein routing the standardized representation of the optimization request based on the characteristics of the optimization request comprises:
accessing a set of mappings that map the characteristics of the optimization request to a cluster of optimizers; and
routing the standardized representation based on the mappings.
9 . The computer implemented method of claim 1 wherein routing the standardized optimization comprises:
routing the standardized representation to a cluster of optimizers based on the type information.
10 . The computer implemented method of claim 9 wherein running an optimization algorithm comprises:
running a different optimization algorithm with each optimizer in the cluster of optimizers to generate a different optimization result with each of the different optimization algorithms.
11 . The computer implemented method of claim 10 and further comprising:
generating an optimization result with a generative artificial intelligence (AI) system based on the standardized representation.
12 . The computer implemented method of claim 11 wherein generating an output indicative of the optimization result comprises:
selecting, from the different optimization results generated by the different optimization algorithms and generated by the generative AI system, a set of the proposed optimization results; and
generating the output indicative of the selected set of optimization results.
13 . A computer system, comprising:
at least one processor;
a first optimizer, implemented by the at least one processor, configured to run a first optimization algorithm, based on information in an optimization request, to generate a first optimization result;
a second optimizer, implemented by the at least one processor, configured to run a second optimization algorithm, based on information in the optimization request, to generate a second optimization result;
a generative artificial intelligence (AI) system configured to receive user input information indicative of the optimization request by conducting a chat user experience with a generative AI system, extract type information from the user input information, the type information being indicative of a goal type and a constraint type of the optimization request, and to generate a standardized representation of the optimization request, based on the user input information, the standardized representation including the type information; and
an optimization router configured to route the standardized representation of the optimization request to the first or second optimizer based on the type information, wherein the optimization router consults mappings that map the optimization goal type and the constraint types in the standardized representation to a cluster of optimizers.
14 . The computer system of claim 13 wherein the first optimizer comprises:
a first cluster of optimizers, each optimizer in the first cluster of optimizers running a different optimization algorithm.
15 . The computer system of claim 14 wherein the second optimizer comprises:
a second cluster of optimizers, each optimizer in the second cluster of optimizers running a different optimization algorithm.
16 . The computer system of claim 15 wherein optimization router comprises:
a parsing system configured to parse the standardized representation to obtain characteristics of the optimization request; and
a cluster selection system configured to route the standardized representation of the optimization request based on the characteristics of the optimization request.
17 . The computer system of claim 16 wherein the cluster selection system comprises:
a classifier configured to receive the characteristics of the optimization request as classifier and to generate, as a classifier output, a cluster identifier identifying the first or second cluster of optimizers based on the inputs to the classifier.
18 . The computer system of claim 16 wherein the cluster selection system comprises:
a rules-based cluster identification system configured to access a set of mappings that map the characteristics of the optimization request to a cluster of optimizers and route the standardized representation to the first cluster or the second cluster based on the mappings.
19 . A method, comprising:
conducting a chat user experience with a generative artificial intelligence (AI) system to obtain information defining an optimization request;
extracting, by the generative AI system from the chat user experience, characteristics of the optimization request from the information, the characteristics comprising a goal type that indicates a type of optimization goal and a constraint type that indicates a type of constraint in the optimization request;
routing the optimization request to an optimizer based on the information defining the optimization request, wherein routing is performed by consulting mappings that map the optimization goal type and the constraint types in a standardized representation to a cluster of optimizers; and
generating an optimization result with the optimizer based on the information defining the optimization request.
20 . The method of claim 19 wherein conducting a chat user experience comprises:
extracting, with the generative AI system, from the information defining the optimization request, type information indicative of a type of optimization request; and
generating a standardized representation of the optimization request, based on the type information, wherein routing the optimization request comprises routing the optimization request based on the type information in the standardized representation.