Robobob

Introduction

Robobob is a fun and interactive command-line-based math robot designed for children. It helps solve arithmetic problems and can also answer basic questions about itself, such as its name.

Get started

Build the Docker Image

To build the Docker image, navigate to the directory containing the Dockerfile and run the following command:

docker build -t robobob .

This command creates a Docker image named robobob. Since some AI libraries need to be downloaded, the process may take a few minutes.

Run the Game

Once the image is built, start the container in interactive mode by running:

docker run -it robobob

-it: Enables interactive terminal mode, allowing you to interact with the game directly in the terminal.

Local Build

If you’d like to build and run the project locally, make sure your system has the following prerequisites installed:

• Java 17
• Maven 3.8.4
• Python3
• spaCy (for Python NLP capabilities)

Design

Framework

• Spring Boot
• Spring Shell

Core Entities

• AnswerProvidable: This interface defines how questions are answered. Once a question is asked, an appropriate implementation of this interface is selected to provide an answer.
  • answer(String question): Provides the answer to the question.
  • belongsToThis(String question): Determines whether the current question can be handled by this answer provider.
• AnswerProviderFactory: Implements the Factory Pattern. It takes the user question as input and returns the most suitable AnswerProvidable implementation.
• SimilarityService: Calculates the semantic similarity between user-asked questions and predefined questions to find the closest match.
• PredefinedQuestion: Stores predefined questions and answers. These questions are persisted in a JSON file and loaded into memory when the game starts.

Workflow

1. Before the game starts, all predefined questions are loaded via the PredefinedQuestion class.
2. When a question is asked, the AnswerProviderFactory selects the appropriate answer provider.
3. Two answer providers are currently supported:
   • ArithmeticAnswerProvider: Handles mathematical questions.
   • PredefinedAnswerProvider: Finds the most semantically similar predefined question and provides the corresponding answer. The similarity is calculated using a Python script powered by the spaCy NLP library. If no match is found, it responds with “Unable to answer.”
4. The game is then ready to process the next question.

Extensibility

Database access

Currently, predefined questions are stored in a JSON file. In the future, if we want to store these questions in a database, it can be done easily by implementing a PredefinedQuestionRepository interface for database access.

Adding a New Predefined Question

Predefined questions are stored in a JSON file and can be easily updated. To add a new question, simply append a new JSON object with the question and answer fields.

Adding a New Answer Provider

To add a new answer provider:

1. Implement the AnswerProvidable interface in a new class.
2. Mark it as a Spring-managed bean.

Once completed, the new provider will be automatically detected by AnswerProviderFactory and ready to answer questions.

Challenges

• Finding the optimal threshold for determining question similarity has been challenging.
• Performance improvements are needed to handle and process user questions more efficiently.
• Implementing machine learning models, along with a feedback loop, can help improve accuracy by allowing the system to better understand user queries and continuously refine its responses based on user feedback.