enterprise approach for creating futuristic NLP-based telephonic / voice chatbots using zero-shot learning

  • Firstly user will call a phone number powered by a communication API service such as voyage or Azure Telephony or in house network
  • Communication API service will first authenticate user basis phone number and in case TPIN and other details are matching will authenticate the user and return a JWT or token depending on the kind of backend
  • Communication API service will also convert user-requested voice from speech to text and pass it to the customer self-service application
  • In case the bot is not able to solve query for the user or if NCCO(Call Control Object) returns to default callback multiple times application will connect the user with a real user sitting in a call center if it’s open else prompt that someone from their team will call customer soon as per timings of call center
  • Customer self-service application will pass NCCO(Call Control Object) to bot logic for NLP processing which can be taken care of RASA AI or custom Zero-Shot Learning NLP engine or other engines like bot framework, Dialogflow etc
  • Once intent and entities are parsed, the parsed requests will be sent to API for further processing. Bot logic will identify intent and entity and call the corresponding microservice or extract data from ETL to create a response statement which could be dynamic or static
  • Small Talk+ Dynamic statements e.g. response to Hi could be Hello, how are you doing? OR Hello FIRST_NAME where the first name is dynamic. Other queries example block my credit card, where bot responds which one- ending with 8875 or 9987?
  • Third-party services like integration with other APIs or banking services. If there is data coming from multiple sources use secured Microservices to fetch dynamic data and build dynamic responses
  • The BOT Logic will interact with ETL which will process structured and unstructured data. 
  • Each request user request and parsed request will also get logged in ELK, which will be used for retraining the BOT model
  • All requests and responses will be over HTTPS following OWASP standards but depending on the configuration we could configure all queries to pass through the firewall or not
  • The dynamic statements depending on whether the response came from small talk or dynamic statement generation from Microservices will be sent back to the customer self-service application
  • Communication API service will convert text back to speech and return to the user
  • All unauthenticated requests will be blocked by the firewall

I am writing after a long time, there could be custom advice and thoughts. Feel free to write back. Love to all. Nikhil

Advertisement

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s