A simple prototype of a chatbot api that can assist with accounting-related tasks.

Create a postgresql instance, create database named attelas then run the db.sql script.

Set openai api key in system env named OPENAI_API_KEY.

you can run with IDE such as Visual Studio or Rider, or run with dotnet CLI with in solution root path.

dotnet build 
dotnet run --project Attelas.WebApi

1. update docker-compose.yml with your OpenAI API Key
2. run with docker compose

docker-compose up -d

curl -X 'POST' \
  'http://localhost:5114/api/chat/completions' \
  -H 'accept: text/plain' \
  -H 'Content-Type: application/json' \
  -d '{
  "text": "Can you provide the status of invoice INV-1020?"
}'

{
  "content": "The status of invoice INV-1020 is \"Pending\".",
  "role": "Assistant"
}

curl -X 'POST' \
  'http://localhost:5114/api/chat/completions' \
  -H 'accept: text/plain' \
  -H 'Content-Type: application/json' \
  -d '{
  "text": "Thanks for sending over the invoice, have a great weekend"
}'

{
  "content": "You're welcome! If you need any help with the invoice or any other accounting-related task, feel free to ask. Have a fantastic weekend!",
  "role": "Assistant"
}

curl -X 'POST' \
  'http://localhost:5114/api/chat/completions' \
  -H 'accept: text/plain' \
  -H 'Content-Type: application/json' \
  -d '{
  "text": "Do we have any invoices due for Acme Corp"
}'

{
  "content": "Yes, we have two pending invoices due for Acme Corp. Here are the details:\n\n1. Invoice Number: INV-1001\n   Due Date: September 30, 2023\n\n2. Invoice Number: INV-1020\n   Due Date: September 30, 2023\n\nIf you need more information or assistance with these invoices, feel free to let me know!",
  "role": "Assistant"
}

curl -X 'POST' \
  'http://localhost:5114/api/chat/completions' \
  -H 'accept: text/plain' \
  -H 'Content-Type: application/json' \
  -d '{
  "text": "What'\''s the current status of INV-1003? I can'\''t find it in the system."
}'

{
  "content": "The current status of invoice INV-1003 is \"Overdue\".",
  "role": "Assistant"
}

curl -X 'POST' \
  'http://localhost:5114/api/chat/completions' \
  -H 'accept: text/plain' \
  -H 'Content-Type: application/json' \
  -d '{
  "text": "Could you check if INV-1020 has been processed?"
}'

{
  "content": "Invoice INV-1020 with the status \"Pending\" has not been processed yet. It is still pending.",
  "role": "Assistant"
}

curl -X 'POST' \
  'http://localhost:5114/api/chat/completions' \
  -H 'accept: text/plain' \
  -H 'Content-Type: application/json' \
  -d '{
  "text": "Could you please remind me address of Acme Corp and submit a new invoice for them please, thanks"
}'

{
  "content": "Sure! The address of Acme Corp is 123 Main Street, Anytown, USA. Please provide me with the details for the new invoice you would like to submit to Acme Corp, including the invoice number, date, amount, and any other relevant information.",
  "role": "Assistant"
}

My solution built on top of the following technologies:

• ASP.NET Core
• OpenAI SDK
• Dapper (An easy to use Object-Relational Mapping toolkit)
• Postgresql (Database)
• Docker (Containerization)

The Clean Architecture is used to structure the project. The project is divided into the following layers:

• Domain: Contains the entities, interfaces, and business logic.
• Application: Contains the use cases and application services.
• Infrastructure: Contains the implementation of the interfaces defined in the Domain layer.
• WebApi: Contains the controllers and the API configuration.

Since this solution is a prototype within several HOURS, it has some known issues:

1. Intent classification and slot filling with LLM(OpenAI in this project) is not working as expected, Prompts need to be repeatedly tested. Joint-BERT or Multi-Turn Conversation might be a possible solution, but it is not in the scope of this project.
2. Authentication/Authorization is not implemented.
3. DI lifecycle was not checked.
4. Error handling is very simple.
5. No unit tests.