Orchestrator

Here is an architectural overview of the Orchestrator Service.

Orchestrator Architecture

Orchestrator

Orchestrator service works as a dynamic tool that simplifies the execution of a bot by transforming an array of transformers into a structured state machine using the Xstate library. This facilitates the seamless coordination and execution of specific functionalities represented by individual transformers within the system. Orchestrator ingests a "logic definition" and creates a state machine out of it, which enables chaining different Transformers together to create a customized bot flow.

API Endpoints

1. /prompt

POST

/prompt Executes bot and sends response to outbound service

Request Body

Name	Type	Required	Description
logicDef	JSON	YES	A bot's logic definition, which is used to create a state machine. Learn more about how to chain multiple transformers to create a logic definition here.
message	XMessage	YES	User's message in XMessage format.
userHistory	Array<XMessageModel>	NO	User's conversation history with a bot. Currently, only last 2 (user's) messages are considered for processing.

Curl Request

curl --location 'http://localhost:3001/prompt' \
--header 'Content-Type: application/json' \
--data '{
    "xstate": [
        {
            "id": "translate",
            "type": "Translate",
            "config": {
                "provider": "bhashini",
                "openAIAPIKey": "vault/openAIAPIKey",
                "bhashiniAPIKey": "vault/bhashiniAPIKey",
                "bhashiniUserId": "vault/bhashiniUserId",
                "userLanguage": "hi",
                "outputLanguage": "en"
            },
            "states": {
                "onSuccess": "llm"
            }
        },
        {
            "id": "llm",
            "type": "LLM",
            "config": {
                "model": "gpt-4",
                "bhashiniURL": "vault/bhashiniURL",
                "outboundURL": "vault/outboundURL",
                "temperature": 0,
                "openAIAPIKey": "vault/openAIAPIKey",
                "bhashiniAPIKey": "vault/bhashiniAPIKey",
                "bhashiniUserId": "vault/bhashiniUserId",
                "outputLanguage": "hi"
            },
            "states": {
                "onSuccess": "done"
            }
        }
    ],
    "message": {
        "messageType": "TEXT",
        "messageId": {
            "Id": "985adbc6-8481-4d85-89e0-1bf0cd131692",
            "channelMessageId": "985adbc6-8481-4d85-89e0-1bf0cd131692"
        },
        "to": {
            "userID": "5b1d9f5b-62d3-4e2b-93fd-9c46c94d2d15"
        },
        "from": {
            "userID": "81ac6cb0-5956-4483-94c6-a7882fc723f5",
            "meta": {
                "mobileNumber": "9999999999"
            }
        },
        "channelURI": "Whatsapp",
        "providerURI": "Gupshup",
        "timestamp": 0,
        "messageState": "REPLY",
        "payload": {
            "text": "गेहूं उगाने का सबसे अच्छा समय?"
        }
    },
    "userHistory": []
}'

In the above example we are executing a bot which uses 2 tranformers (Translate and LLM).

Where this bot flow works as below:

(user)[question in hindi] -> (bot)[translates user message from hindi to english] -> (bot)[sends this to LLM and gets response] -> (bot)[translates back to hindi] -> (bot)[sends response to user]

Above example is just an simple translate+LLM execution but, One can create any type of custom flow combining various transformers available.

---