All Classes and Interfaces
Class
Description
Action Risk , all medium risk aciton will need one human verification, high will need two
All Types of actions implemented so far
use this annotation to mark the loader as active
This is the key class in creating all the AI related action.
Marker interface for executors
Multi Cloud AI platform
This Action can be used when you want the AI to get back with answer and if you do not have any
specific action to be performed.
Base class for AI Hallucination and Bias detection
The
ExplainDecision interface represents a mechanism for AI to explain decisions
regarding a particular prompt text, method name, and reason.Shell, HTTP and Java Methods are supported by default out of the box most of the tasks could be accomplished by the same
however there might be a situation or use case to create custom set of actions *Please note* this is different from Custom actions.
The main processor class, can execute single action or multiple action in sequence Uses Gemini Function calling
- based on prompt can predict and trigger action
- based on prompt can predict and trigger multiple actions sequentially
- based on prompt can predict and trigger multiple actions parallely
- Can take HumanInLoop object and wait for Human Validation
- Can take ExplainDecision Object and provide why the decision was taken by AI
Uses Gemini to check the user prompts
Take actions based on images
This class takes a prompt and can build Java Pojo out of it, it could also transform the prompt into json with
name and value
so if you pass it a json like this
String jsonString = "{\"lastName\":\"String\",\"firstName\":\"String\",\"reasonForCalling\":\"String\",\"dateJoined\":\"date\",\"location\":\"String\"}";
and then pass a prompt
I went to the part yesterday and met someone it was so good to meet an old friend.
This class takes a prompt and can build Java Pojo out of it, it could also transform the prompt into json with
name and value
so if you pass it a json like this
String jsonString = "{\"lastName\":\"String\",\"firstName\":\"String\",\"reasonForCalling\":\"String\",\"dateJoined\":\"date\",\"location\":\"String\"}";
and then pass a prompt
I went to the part yesterday and met someone it was so good to meet an old friend.
If guard rails are broken
Interface for checking guard rails
This is a Generic Http Action Class , it will be built from the config file.
The
HumanInLoop interface represents a mechanism for allowing human involvement
in a feedback loop process.Maps to Input parameter in the config file, this is the input which will be sent to HTTP request
This is the base class for all the java based executors with common functionality
Create a POJO Object from the response object
This is one of the main classes which is part of processing logic.
Simple implementation of Human IN Loop, all it does is logging but in real world this could trigger a full
human validation and return true or false
Uses Json conversion to convert method and java pojo to jsons and then call the openai , your objects have complex
parameters like custom dates etc then its better to pass a custom GSON , otherwise it will use the default gson
Just a marker interface to denote that the implementing classes will be predicted by default.
The
PredictionLoader class is responsible for managing the prediction process
by interacting with various prediction models and loading actions based on predictions.Processes action script line by line , previous result is validated
Base class to execute shell commands , these configuration are part of shell_action.yaml.
The
ShellPredictionLoader class is responsible for loading shell actions from a YAML file
and populating a map of predictions within the application.This will ensure that the action classes are loaded from Spring Applicaiton Context rather than
creating the new one , the advantage of that is we can maintain spring dependency injection for all the beans
Uses Gemini for processing
This will ensure that the action classes are loaded from Spring Applicaiton Context rather than
creating the new one , the advantage of that is we can maintain spring dependency injection for all the beans
Uses OpenAI for processing
This class is Zero Shot method designed to assess the consistency of responses generated by a Large Language Model (LLM) and detect potential hallucinations.