VOYAGER

Voyager specially build for Pepper Robot Chatbot. Check also pepper-chatbot.

BUILD & RUN

This application has interacting with AWS Resources, and it's build specificially inside an EC2 or ECS, if using outside the environments you might need to install aws-cli to config the identifications.

Make sure you have installed docker, docker compose and make on machine unless you are using containers like ECS.

To build and run, simply run command

make up
# OR
make dev # for development setup. Using this command also requires run
npm install pnpm && pnpm install # To install dependencies
# Also you can use any package manager you perfer

Routes

Healthy

Route

/healthy

Description

The route to check container healthy status

Example Usage

GET /healthy HTTP/1.1

Example Response

ok

Chat Completions

Route

/v1/chat/completions

Description

The route for inference connected to AWS Bedrock service. Request and response in OpenAI format.

Example Usage

POST /v1/chat/completions HTTP/1.1
Content-Type: application/json

{
    "messages": [
        { "role": "system", "content": "You are an assistant named Pepper helps users answer their questions" },
        { "role": "user", "content": "Hello" }
    ],
    "max_tokens": 50,
    "model": "anthropic.claude-3-sonnet-20240229-v1:0"
}

Example Response

{
  "id": "<random-id>",
  "object": "chat.completion",
  "created": 1728021450456,
  "model": "anthropic.claude-3-sonnet-20240229-v1:0",
  "system_fingerprint": "<random-fp>",
  "choices": [
    {
      "index": 0,
      "message": [
        {
          "role": "assistant",
          "content": "Hi!"
        }
      ],
      "logprobs": null,
      "finish_reason": 'stop'
    }
  ],
  "usage": {
    "prompt_tokens": 0,
    "completion_tokens": 0,
    "total_tokens": 0
  }
}

Notes

If the request messages contains only user and system, the code treats the messages as new conversation and will rebuild the AWS Bedrock Client.

Speech Recognition

Route

/v1/speech/recognition

Description

The route for speech recognition, backend is Azure Speech Services, requires key and region of Azure. See Azure Speech Recognition part.

Example Usage

POST /v1/speech/recognition HTTP/1.1
Content-Type: audio/wav

# <audio buffer which content is "hello world">

Example Response

{
  "text": "hello world!"
}

If there are any error occurs, you will receive

{
    "error": <the error object>
}

Azure Speech Recognition

To set up the azure speech recognition service, you need to create your own service on Azure portal and your key and region ready.

Create a file named .env.azure and fill your key and region:

AZURE_SPEECH_KEY=<your-aws-speech-sercive-key>
AZURE_SPEECH_REGION=<your-aws-speech-sercive-region>

The app will load it automatically.