Ultimate AI

Ultimate AI is an application to serve customers as a chat bot.

Technolojies Used

• Spring Boot 2
• Java 8
• Maven 3
• Mongo Db Atlas
• OpenApi - Swagger
• Lombok
• Mockito

Requirements

• Maven 3
• Jdk 8

Test and Run

Run the following command to test and run the application;

cd ultimate-ai && mvn clean install && cd ultimate-ai-app && mvn spring-boot:run

Or run the main class on an IDE after maven clean install(generated code is not commited).

Access

Chat bot can be accessed on;

• http://localhost:9090/swagger-ui.html#

Or post can be applied directly on;

• http://localhost:9090/chat-bot

Clean Code Actions

• Annotations are used to keep the code simple.
• An api module is created to keep the generated code isolated and manage dependencies seperately.
• IntentApiService is written in api module to prevent generated code to cause test failures in app module.
• Constants are located in the same class on purpose to prevent a big mess constant class.
• Sonarlint is used and issues are handled.

Choices and Reasons

• For database, I came up with two solutions as an online database or Docker Compose with a local one. I used Mongo Db Atlas to be able to connect from anywhere. It was a simpler and lighter solution.
• For confidence treshold, I thought what matters is how better is the best intent from the second intent, the other intents makes no difference. I observed that when I send "goodbye" as request, AI returns "Thank you" with confidence 1 and "Goodbye" with confidence 0.81. So I defined a configurable fallibility percentage and set it to 20%.

What Could Be Done More

• ApiKey could be encrypted an passed as an argument.
• BotId pattern could be validated for differentiated error messages.
• Controller could be tested as mvc in case of http status of the response matters.
• In a bigger project, interfaces of the components could be used as contract.
• In a bigger project, -data , -service, -web modules would be created to separate dependencies and test needs.