Introduction

Overview of Kua.ai

Welcome to Kua.ai, an innovative content creation tool tailored for the e-commerce sector. This platform is designed for a diverse array of users, including online store owners on Shopify and WooCommerce, sellers on Amazon, e-commerce departments in consumer brands, and content creators or bloggers on WordPress.

Kua.ai stands out as a tool that simplifies the process of generating factual, on-brand, and SEO-optimized content, thereby streamlining content production for platforms like Shopify, WordPress, and Amazon. It reduces manual effort and facilitates content automation, making it a vital asset for enhancing your e-commerce presence.

Target Readers of this Doc

• AI Application Developers: These are individuals with the ability to abstract e-commerce challenges into AI solutions. They are familiar with the fundamental mechanisms of Kua.ai's AI Applications and can construct applications based on specific requirements. In real life, they can be an entrepreneur or business leader with experience in Gen AI, or an AI consultant in a business or agency who can build applications for their clients.
• General Users: This group includes users who engage with the AI applications developed by the AI Application Developers. They utilize these applications to fulfill specific functions and needs, understanding the problems solved by the AI applications, including their inputs and outputs. Below are examples demonstrate how general users from different backgrounds can effectively use AI applications on Kua.ai to enhance their e-commerce activities and content strategies.
  • Product Description Generation: A Shopify store owner can use an AI application to automatically generate unique and compelling product descriptions, saving time and improving the SEO of their online store.
  • Market Trend Analysis: An Amazon seller can leverage an AI application to analyze market trends and consumer behavior, helping them make informed decisions about inventory and marketing strategies.
  • Content Creation for Blogs: Bloggers on WordPress can use an AI application to generate ideas and draft initial content for their blogs, especially for topics related to e-commerce and consumer products.
  • Social Media Content Automation: A brand's e-commerce department can utilize AI applications to create and schedule social media posts, ensuring consistent online presence and engagement with customers.

Purpose of the User Manual

This manual is designed to provide comprehensive guidance on utilizing Kua.ai, catering to both AI application developers and general users. Whether your goal is to develop advanced AI applications or simply use existing applications for specific e-commerce functions, this manual will guide you through every step, offering valuable insights to maximize your experience on the Kua.ai platform.

Application vs AI Chats

Kua.ai provides two primary content creation interfaces: Chat and Applications. Chat is an AI-powered interface that interacts with users to answer questions and generate content as per the users' requests, making it an accessible and interactive option for single-user content creation.

On the other hand, Applications is built for more extensive tasks, offering a more standard and automated way to complete tasks. From business perspective, it pre packages the best practices for completing certain tasks so the results are more reliable. For example, when tasked with creating an Amazon Listing, the application mandates the inclusion of keywords and stipulates that the title and key features comply with specific rules regarding length, formatting, and structure. This ensures that the Amazon Listing generated by the application better meets the success criteria. From a technological perspective, it features an innovative technique known as prompt engineering and chaining for generating complex content.

In essence, while the Chat interface caters to the needs of individual users with its user-friendly and interactive design, Applications is better suited for organizational needs, facilitating the creation of substantial content and task automation. Both interfaces are crafted to be efficient and accurate, serving a wide range of content generation needs.

Understanding Kua.ai for AI Application Developers

Kua.ai is a platform that empowers developers to create AI Native Applications tailored for various e-commerce scenarios. Here, we introduce the fundamental concepts and mechanisms that form the backbone of application development on Kua.ai.

Core Concepts

• Applications: These are the primary units of functionality within Kua.ai. Applications are designed to perform specific tasks using AI.
• Functions: Functions are the building blocks of applications. They encapsulate specific AI capabilities and can be combined in various ways to create complex applications.
• Workflows: Workflows define how functions interact and work together within an application. They determine the sequence and conditions under which functions execute.

Function Types

Native Functions

These are pre-built functions provided by Kua.ai, offering standard AI capabilities.

How to access the native functions in Kua.ai

You can have access to the native functions in two ways: use them in a semantic function or in a workflow function.

Use native functions in a semantic function

• Go to the section of "Build an App"

• After completing the basic description of an App, click next, and select the "Prompt Mode" tab. On the bottom left you can see the "Select Tool" drop down window.

• You can see all the available native functions in the drop down window. Select the one you need and the native function you selected will be used in the prompt via function call.

Use native functions in a workflow function

• Go to the section of "Build an App"

• After completing the basic description of an App, click next, and select the "Workflow Mode" tab

• Click the "..." on the left of an action (function) within a workflow, and then click on the "Change" button, a new window will pop up.

• You can see all the available native functions in the pop up window, within the Built-in Apps channel (on the left). You can select the one you need and click "Confirm" to use it in your workflow.

List of native functions available as of Jan 19, 2024

1. Get current-time: This function enables GPT to retrieve the current time.

2. google-search: Provides GPT the capability to perform a Google search based on a user's query.

   a. Input: a search query or a keyword b. Output: the top 10 google search results (titles, snippets or URLs)

3. google-search-related-searches: This function enables GPT to retrieve Google's related search queries.

   a. Input: a search query or a keyword b. Output: the related search terms

4. google-search-also-ask: Designed to allow GPT to access the "People Also Ask" section in Google search results.

   a. Input: a search query or a keyword b. Output: what people ask for the search term

5. keywords search volume trend: It allows GPT to obtain search volume trends for specified keywords, taking into account various parameters like country and data source.

   a. Input: i. a list of search queries or keywords. IMPORTANT NOTE about the input format: 1. You should use commas to separate each query, following the format "query 1, query 2, query..." 2. If the function utilizes the output of a preceding function as input, it is advisable to specify the format of the previous function accordingly(see below example).

   

   ii. The country or market where the data is relevant, e.g. Global, United States, Canada, etc. iii. Data source, you can either enter "gkp" (google keyword planner) or "cli" (gkp + clickstream data) iv. Currency: the currency for the CPC. You can enter "usd" for US dollars.

   b. Output: search volume for the list of queries in the past 12 months, and the data source is Google.

   

6. keywords search volume: This function helps GPT to determine the search volume, CPC, and competition level for specified keywords. a. Input: i. a list of search queries or keywords. IMPORTANT NOTE about the input format: 1. You should use commas to separate each query, following the format "query 1, query 2, query..." 2. If the function utilizes the output of a preceding function as input, it is advisable to specify the format of the previous function accordingly(see below example).

   

   ii. The country or market where the data is relevant, e.g. Global, United States, Canada, etc. iii. Data source, you can either enter "gkp" (google keyword planner) or "cli" (gkp + clickstream data) iv. Currency: the currency for the CPC. You can enter "usd" for US dollars.

   b. Output: search volume, CPC (cost per click) and competition for the list of queries.

   

7. keywords search volume only: Similar to the above, but focused solely on providing the search volume of keywords. a. Input: i. a list of search queries or keywords. IMPORTANT NOTE about the input format: 1. You should use commas to separate each query, following the format "query 1, query 2, query..." 2. If the function utilizes the output of a preceding function as input, it is advisable to specify the format of the previous function accordingly(see below example).

ii. The country or market where the data is relevant, e.g. Global, United States, Canada, etc. iii. Data source, you can either enter "gkp" (google keyword planner) or "cli" (gkp + clickstream data) iv. Currency: the currency for the CPC. You can enter "usd" for US dollars. b. Output: search volumes for the list of queries.

8. Amazon product information Tool: Designed to enable GPT to fetch basic information about products on Amazon using their ASINs. a. Inputs: 1. the Amazon ASIN 2. The marketplace where the ASIN is active on, e.g. amazon.com, amazon.de, amazon.ca, amazon.es, amazon.fr, etc. b. Output: detailed data related to that ASIN, title, key features, product description, A+ content, specs, review rating, number of reviews, image/ video URL, etc.
9. amazon-product-search: Allows GPT to search for Amazon products using specific search terms and parameters. a. Inputs: 1. the Amazon ASIN 2. The marketplace where the ASIN is active on, e.g. amazon.com, amazon.de, amazon.ca, amazon.es, amazon.fr, etc. 3. sort_by: the ranking order of search results. valid values are: featured; most_recent; price_low_to_high; price_high_to_low; aeverage_review. 4. exclude_sponsored: whether to exclude sponsored ASIN/listing. The default is "no". 5. Limit: Number of search results (ASINs) returned b. Output: top N listings/ ASINs as the search results, together with their titles, URLs, etc.
10. ABA Keyword Ranking: This function helps in obtaining popularity (search frequency rank) information based on ABA (Amazon Brand Analytics) data. Currently, this data is only available for amazon.com. a. Input i. Keywords: a list of search queries or keywords. IMPORTANT NOTE about the input format: 1. You should use commas to separate each query, following the format "query 1, query 2, query..." 2. If the function utilizes the output of a preceding function as input, it is advisable to specify the format of the previous function accordingly(see below example).

ii. Format: format of the output - it can either be jason or markdown. b. Output: Search Frequency Rank for the list of keywords/ search queries in two period - last week, and the week before last week. 11. ABA Ranking Change. This function helps in obtaining popularity (search frequency rank) change information based on ABA (Amazon Brand Analytics) data. A sudden rise in popularity also means sales opportunities. Currently, this data is only available for amazon.com. a. Input i. Keywords: a list of search queries or keywords. IMPORTANT NOTE about the input format: 1. You should use commas to separate each query, following the format "query 1, query 2, query..." 2. If the function utilizes the output of a preceding function as input, it is advisable to specify the format of the previous function accordingly(see below example).

ii. Format: format of the output - it can either be jason or markdown. b. Output: Search Frequency Rank change for the list of keywords/search queries is defined as the rank from the last week compared to the rank from 2 weeks earlier. A negative number indicates a 'drop' in rank, while a positive number indicates a 'rise'.

Semantic Functions

Semantic functions are user-defined and are tailored to specific use cases, allowing for greater flexibility and customization.They are a pivotal aspect of AI application development in Kua.ai, offering a high degree of customization and precision. Here’s an overview of their key components:

• Prompt Template: A SemanticFunction defines a Prompt Template, which becomes a prompt for a Large Language Model (LLM) after combining parameters. Parameters are integrated using placeholders like {{ parameter }}. Such as {{Keyword}} in the below example.

• LLM Agent Context: This includes configurations related to the invocation of the underlying LLM in an agent mode. Components of the LLM Agent Context are:
  • LLM Settings: Configurations like the model name and temperature setting.
  • System Prompt: The initial prompt or instruction given to the LLM.
  • Language and Writing Style: Settings to specify the language and the desired style of the generated content.
• LLM Configuration: Essential configuration information required to invoke the LLM, such as the model URL and the OpenAI API access key.
• Tools: These are encapsulations provided for LLM invocation, enhancing the functionality and flexibility of the SemanticFunction.
• Knowledge Base: A collection of knowledge items, enabling SemanticFunctions to use Retrieval-Augmented Generation (RAG) by referencing the KnowledgeBase. This allows for more context-aware and relevant content generation. A typical feature in Kua.ai using Knowledge Base is the Infobase function, where a user can save their knowledge and use them in either semantic or workflow functions.

Workflow Functions

These functions manage and orchestrate the flow of data and operations within an application.

• Default Output Behavior: By default, the output of a WorkflowFunction is determined by the last executed Function node within it. However, this behavior can be customized.
• Custom Output Combinations: Developers have the flexibility to define a Function node that combines the outputs of each Function node within the WorkflowFunction. This custom node then serves as the workflow's final output.
• Function Output Types: Referring to the standard Function definition $$Function: (x: X) => y: Y$$, the output type of a Function follows the constraints set by its definition. This consistency is crucial for the integrity of the workflow.
• Syntax for Output Definition: To facilitate developers in defining the expected output of a WorkflowFunction, a syntax enhancement is provided. This 'Output' syntax aids in combining results from different Function nodes in the workflow. Essentially, defining an 'Output' is equivalent to declaring the output type of the WorkflowFunction and creating a Function node to construct the value of this output type.
• Language and Subclass Features: In Kua.ai, 'Workflow' typically refers to WorkflowFunction. This subclass of Function extends capabilities for constructing workflows, including:
  • The ability to combine multiple Functions to construct a WorkflowFunction, with the capability for one Function's output to be referenced by another.
  • WorkflowFunction requires explicitly defined input parameters and an output format/type.

Function Inputs

In addition to the core concepts of applications, functions, and workflows, it is crucial for AI application developers to understand the concepts of Function Parameters and Arguments within Kua.ai.

Function Parameters

• Definition: Function parameters are the input declarations or definitions for a function. They specify the types and nature of the parameters that the function accepts.
• Usage in Applications and Workflows: When calling a function, either within an application or when building a workflow, it's necessary to pass the corresponding arguments as per the defined parameters.

Function Arguments

• Definition: Function arguments are the actual values provided as inputs when calling a function.
• Example: In a function definition, a parameter could be labeled as asin. When this function is called, the specific value (e.g. B0CQJZC5R1) provided for asin is the argument.

Input of Promt Application

When building applications in prompt mode (specifically, applications of the semantic function type), you can define the input for an application by using {{Parameter}}, where 'Parameter' represents the name of your parameter. For instance, in the example provided, the parameters are 'Title' and 'Key Feature'. Thus a general user needs to fill out these outputs in order to run the application.

Input of a Workflow Application

When building a Workflow Application, it is essential to specify the input required for the application. This process resembles that of a prompt application. You should enter the names of your parameters and provide a description in the name section. This description will offer general users hints on how to input the necessary information.

Input of a step in a workflow

Additionally, when constructing a workflow application, it's necessary to define the input for each step. You have the option to use text as inputs, or you can enter '#' to view the available arguments. These arguments could be an input from the workflow, information stored in your infobase, or outputs from previous steps within the same workflow application.

Function Outputs

Function Output Types and Document Generation

Kua.ai enables developers to create functions that output specific types of documents, such as Amazon Listings, Blogs, Tweets, etc. This means a function is designed to produce a particular type of document, integral to various workflows.

For example, consider a Twitter document structure - a function whose output type is Twitter, like generateTwitter(topic: String): Twitter. Such a function is specialized for generating Twitter-type documents and can be integrated into Twitter operation workflows, allowing users to generate Twitter posts automatically.

Relationship Between Functions and Applications

• Functions: These are developer-oriented concepts, exposing abstract and complex features. They are designed for rigorous and efficient development.
• Applications: These are user-facing concepts, the foundation of using Kua.ai's AI capabilities. They are designed with a focus on ease of use and accessibility.

Developers create and maintain Functions. These functions, once supplemented with user-oriented descriptions and help information, are published as Applications for user consumption.

Output Options

At the end of a workflow application, AI application developers must define the outputs that the application should generate, ensuring they are meaningful to the user. There are two types of output formats to choose from when configuring a workflow application: Action List or Template.

Action List Output

If you choose this option, you will need to specify which steps' outputs will be used in the final output. You can view all the outputs from the workflow steps and select one or multiple of these for the final output.

Template Output

In the template option, you can structure the output in the form of a report. Additionally, you can utilize Markdown to customize the output, making it more readable and easier to understand.

Another benefit of the template option in the output is that it allows you to embed one workflow application with the template output into another. For example, consider the 'Blog Post Workflow_V2', a workflow application that generates a blog. I can build an additional workflow application on top of it and include other steps. In this case, these steps involve revising the blogs by incorporating case studies.

Step by Step Mode

In both of the above output options, AI application developers have the option to allow or restrict general users from viewing the detailed steps of a workflow application by toggling the button below on or off.

Generation (History) of Content

Each content generation by an Application is a 'Generation,' containing inputs and outputs. Generations are categorized by applications and can be associated with specific documents.