Here is information about the md-tab component, that is already installed under @material/web
https://github.com/material-components/material-web/blob/main/docs/components/tabs.md

Copy:
apps/policy-synth/src/ps-raw-evidence.ts

To:
apps/policy-synth/src/ps-root-causes.ts

And modify it to subclass PsRawEvidence and delete unused methods and create new as needed.

The common categories of root causes from the search GPT-4 scanning prompt need to be added.

Task Description
We want to create a new typedef called PsModelChatItem to replace {role:string,message:string}`,

Task Instructions

1. Add new interface PsModelChatItem {role:string,message:string} to src/streamingLlms.d.ts
2. Replace {role:string,message:string} with PsModelChatItem in src/models/baseChatModel.ts and src/models/openAiChat.ts, src/models/claudeOpusChat.ts and src/models/googleGeminiChat.ts

Here are two relevant test prompts:
https://chat.openai.com/share/a015bc27-f3e7-42c8-be7d-507cc58c6168
https://chat.openai.com/share/bc62c60f-245f-4d72-9ccd-fd5938e53a65

Here is the evidence search query GPT-4 prompt (in the format the new prompts needs to be in)
https://platform.openai.com/playground/p/qcsfzJ2zNqJKNDa3nhqLosrh

And example GPT-3.5 16k for scanning a website (you'll need to copy paste some text into :
https://platform.openai.com/playground/p/KewOzWddGXxXCjoQPEEOGHeY

Task Description
Use modelName for Tiktoken token counts in the LLM models classes and add maxTokensOut to base model and constructors of the sub classes

Task Instructions

1. Change src/models/baseChatModel.ts to add a maxTokensOut string property, as number = 4096 and should be set in the constructor and default to 4096 but the users of sub classes must be able to change it
2. Change src/models/clauseOpusChat.ts to use maxTokensOut and this.modelName for the Tiktoken encoding_for_model function instead of cl100k_base (as this does not work for gpt-4o and possibly future models)
3. Change src/models/azureOpenAiChat.ts to use maxTokensOut and this.modelName for the Tiktoken encoding_for_model function instead of cl100k_base
4. Change src/models/openAiChat.ts to use maxTokensOut and this.modelName for the Tiktoken encoding_for_model function instead of cl100k_base
5. Change src/models/googleGeminiChat.ts to use maxTokensOut and this.modelName and keep on using the built in Gemini token count method

Use the getRawEvidence in api/controllers/projectsController.ts as a template for this new API call

Task Description
We want to add a new model class into src/models/azureOpenAiChat.ts using the @azure/openai npm module.

Task Instructions
Add a new model class in src/models/azureOpenAiChat.ts using the @azure/openai npm module based on src/models/baseChatModel.ts parent class.

Task Description
Refactor the baseAgent class for smaller and clearer functions, especially all the JSON parsing in callLlm but also other best practices. Consider if splitting out some functionality into smaller classes, then do that but do not split callLlm into a separate class. The child classes should not have to change and all the code should work exactly the same.

Task Instructions

1. Refactor the src/models/baseChatModel.ts class but not in a way that requires changes to child classes the extend this class.

Copy the two files from here agentQueue/src/agents/policies/ranking to to /src/problems/ranking and override the logic.

Evidence rating prompt example:
https://platform.openai.com/playground/p/j5qtZn7LLGukqscgBsD8norb?model=gpt-4

Task Description
Create LLM classes for OpenAI, Claude Opus and Google Gemini, with a design that allows easy addition of other models in the future. This is a typescript based es module NodeJS modern server application.

Task Instructions
1. Create a new base chat class src/models/baseChatModel.ts in a new file.
2. Then create src/models/openAiChat.ts, src/models/claudeOpusChat.ts and src/models/googleGeminiChat.ts for chat only
3. For the cloudeOpus use the @anthropic-ai/sdk npm
4. For the googleGemini use the @google/generative-ai npm
5. For the openAi use the openai npm
6. The baseChatModel and the child classes should only implement those two methods:
generate([{role:string,message:string}], streaming: boolean | undefined, streamingCallback: Function | undefined)
getNumTokensFromMessages([{role:string,message:string}])

Task Description
Update src/models/baseChatModel.ts, src/models/clauseOpusChat.ts, src/models/azureOpenAiChat.ts and src/models/openAiChat.ts to use this.modelName

Task Instructions

1. Change src/models/baseChatModel.ts to add a modelName string attribute string | TiktokenModel and should be set in the constructor
2. Change src/models/clauseOpusChat.ts to default to "claude-3-opus-20240229" for this.modelName
3. Change src/models/azureOpenAiChat.ts to default to "gpt-4o"
4. Change src/models/openAiChat.ts also to default to "gpt-4o"
5. Change src/models/googleGeminiChat.ts to default to "gemini-pro" for this.modelName in its constructor but use can provide their own, same for the other models

Those two files:
agentQueue/src/agents/policies/web/getEvidenceWebPages.ts
agentQueue/src/agents/policies/web/getRefinedEvidence.ts

Easiest to start by copying them to:
agentQueue/src/agents/problems/web/getRootCausesWebPages.ts
agentQueue/src/agents/problems/web/getRefinedRootCauses.ts

Then remove all the common methods and override methods for the Root Causes logic and data types.

Also copy the evidencExamplePrompts.ts over to the problems/web folder and modify based on your data types with one-shot example for GPT3.5 16k
(hint: You can use GPT-4 in the OpenAI Playground or ChatGPT 4 to create the examples)

Refine prompt example:
https://platform.openai.com/playground/p/hMFPbfsAPafVw5vihG29sqER?model=gpt-4

Task Description
We've created those constructor model options typedefs:
interface PSModelConfig {
apiKey: string;
modelName?: string;
maxTokensOut?: number;
};

interface PSAzureModelConfig extends PSModelConfig {
endpoint: string;
deploymentName: string;
};

interface PSOpenAiModelConfig extends PSModelConfig {
projectId?: string;
};

We need to change the underlying models but we can keep the base class the same do not change it.

Task Instructions

1. Change src/models/clauseOpusChat.ts to use new options instead of function parameters and setup the rest of function in best possible way
2. Change src/models/azureOpenAiChat.ts to use new options instead of function parameters and setup the rest of function in best possible way
3. Change src/models/openAiChat.ts to use new options instead of function parameters and setup the rest of function in best possible way
4. Change src/models/googleGeminiChat.ts to use new options instead of function parameters and setup the rest of function in best possible way

Task Description
Update src/models/clauseOpusChat.ts, src/models/azureOpenAiChat.ts and src/models/openAiChat.ts to use a more effective token count mechanism for the getNumTokensFromMessages method
Task Instructions

1. Change src/models/clauseOpusChat.ts to use something else than just characters, maybe use tiktoken or something?
2. Change src/models/azureOpenAiChat.ts to use something else than calling the api
3. Change src/models/openAiChat.ts also to use something else than calling the api