BFF

BFF is an service that handles all operations that are required by the recipes and admin application. It includes various processes such as document upload, chunk retrieve and recipe config creation.

Recipe Configuration Parameters

Name	Type	Required	Description
llm	JSON	YES	Configuration for LLM transformer
llm.model	String	YES	LLM model to be used in transformer. Eg: gpt-4, gpt-3.5-turbo ..
llm.prompt	JSON	YES	LLM Prompt settings
llm.prompt.system	String	YES	System instructions which needs to be used by the bot while answering user's questions.
llm.prompt.user	String	YES	User instructions for bot, generally context it passed here.
llm.prompt.summarization	String	NO	Specific instructions to llm for summarizing the final answer.
llm.prompt.answerType	String	NO	Type of answer Eg: Descriptive, succinct
llm.enableStream	Boolean	NO	Set this to True in order to stream bot answers instead of getting complete answer at a time.
retriever	JSON	YES	Configuration for Retriever transformer, which is used to get related context to answer user questions.
retriever.documentIds	Array<String>	Yes	List of pdf documents to be used by the bot.
retriever.topK	Number	Yes	Upper limit on how many chunks to retrieve.
retriever.xlsxIds	Array<String>	Yes	List of excel documents to be used by the bot.
translation	JSON	YES	Configuration for Translate transformer
translation.userLanguage	String	YES	Input language of the bot.
translation.outputLanguage	String	YES	Language in which bot should respond back.
translation.provider	JSON	String	Translation service provider Eg: Azure, Bhashini.
audioService	JSON	YES	Configuration for Text To Speech service.
audioService.active	Boolean	YES	Boolean denoting whether to activate text to speech.
audioService.provider	String	YES	Text to Speech service provider Eg: Azure, Bhashini.
audioService.voiceSettings	String	YES	Voice gender. Accepted values: male, female.
audioService.sampleRate	Number	YES	Used to set the speed at which audio is played, default: 8000. higher values results in faster audio play and vice versa.

API Endpoints

1. /bot

POST

/bot Create a bot for RAG recipe

Request Body

{
  "llm": {
    "model": "string",
    "prompt": {
      "system": "string",
      "user": "string",
      "summarization": "string",
      "answerType": "string"
    },
    "temperature": 0,
    "enableStream": true
  },
  "retriever": {
    "documentIds": [
      "string"
    ],
    "topK": 0,
    "xlsxIds": [
      "string"
    ]
  },
  "citation": {
    "style": "string",
    "maxLength": 0
  },
  "translation": {
    "userLanguage": "string",
    "outputLanguage": "string",
    "provider": "string"
  },
  "audioService": {
    "active": true,
    "provider": "string",
    "voiceSettings": "string",
    "sampleRate": 0
  },
  "denseEmbeddingsWeightage": 0
}

Response Format

• Success

{
  "error": false,
  "botId": <uuid>
}

• Failure:

{
    "botId": null,
    "error": true,
    "message": "Error message",
}

2. /bot/:botId

PUT

/bot/:botId Updates an existing bot for RAG recipe

Request Body

{
  "llm": {
    "model": "string",
    "prompt": {
      "system": "string",
      "user": "string",
      "summarization": "string",
      "answerType": "string"
    },
    "temperature": 0,
    "enableStream": true
  },
  "retriever": {
    "documentIds": [
      "string"
    ],
    "topK": 0,
    "xlsxIds": [
      "string"
    ]
  },
  "translation": {
    "userLanguage": "string",
    "outputLanguage": "string",
    "provider": "string"
  },
  "audioService": {
    "active": true,
    "provider": "string",
    "voiceSettings": "string",
    "sampleRate": 0
  }
}

Response Format

• Success

{
  "error": false,
  "botId": <uuid>
}

• Failure:

{
    "botId": null,
    "error": true,
    "message": "Error message",
}

3. /document/upload

POST

/document/upload Upload a document and stores its embeddings.

Query Parameters

Name	Type	Required	Description
format	string	YES	Type of document that is being uploaded. Supported types are: PDF, XLSX, LINK

Request Body

Name	Type	Required	Description
file	binaryFile	YES	csv or an pdf file.

Response Format

• Success

{
    "data": {
        "taskId": "c575540e-c203-419c-8cf1-47ffc6f66ae7"
    },
    "error": false
}

• Failure:

{
    "detail": "Error message",
}

4. /document

GET

/document Retrieves the list of uploaded documents. Supported types = PDF

Query Parameters

Name	Type	Required	Description
page	number	NO	Page number
perPage	number	NO	Number of records to be fetched per page
type	string	YES	Type of document that was uploaded. Supported types are: PDF, XLSX

Response Format

• Success

[
    {
        "id": <uuid>,
        "name": "fileName.pdf",
        "url": <PDF_LINK>,
        "ownerId": <uuid>,
        "orgId": <uuid>
    }
]

• Failure:

{
    "error": "Error message",
}

5. /retrieve

GET

/retrieve Retrieves similar chunks for input query.

Query Parameters

Name	Type	Required	Description
text	number	YES	query against which chunks are retrieved.
topK	number	NO	Upper limit on chunks length.
pdfId	Array<string>	NO	array of documentIds to search from.

Response Format

• Success

[
    {
        "id": <uuid>,
        "content": <string>,
        "heading": <string>,
        "summary": <string>,
        "pdfName": <string>,
        "pdfId": <string>,
        "metaData": <json>,
        "type": <string>,
        "vidio": <link>
    }
]

• Failure:

{
    "error": "Error message",
}

For complete documentation and API execution playground visit -> Swagger