Customer Experience AI (CxAI)

Improve Customer Experience (Cx) by leveraging GenAI capabilities. Uncover hidden business opportunities by understanding customer needs and sentiments. Address customer concerns ahead of time.

Comment

Note about video presentation

The solution was initially implemented used Google Vertex AI Agent but, I realized that I can not get the Chat App approved from Google before the hackathon deadline. The Google approval process need formal privacy policy etc. to successfully complete the approval. Hence, I have added an additional Chat implementation using HTML and JS based chat widget so that Hackathon evaluators and users can access the and test drive the live demo of Chat Agent.

I have recorded a demo of Google Vertex AI Agent App (using my test account) to show what the experience looks like with Vertex AI Agent App. Both videos (using Vertex AI Agent and classic HTML-JS widget) are shared below.

More detail about the approval process: https://support.google.com/cloud/answer/13464323/#exemptions

Video Demo/Pitch and Live Demo:

Please Note: The demo is hosted using a free trial account on (render.com). Free instance will spin down without any activity. Please wait for 50 seconds after your first request for site to come up. Also, since this application is using sqllite, the database is refreshed with every restart of instance.

Inspiration (Business Scenario)

The integration of Generative AI (GenAI) capabilities and tools in any industry holds immense potential to revolutionize the customer experience and streamline operations. Rental Car Industry is one of those industries.

My idea is to leverage Google's low-code/no-code AI solutions to revolutionize the customer experience and streamline operations for rental car business.

An existing rental car company want to leverage AI/ML and GenAI (Google AI, Vertex AI's and Low-Code/No-Code) capabilities to provide exceptional customer experience. The GenAI capabilities will allow customers of the rental car company to complete tasks like booking a car, cancellation etc. with help of GenAI trained Chat Bot/Agent. The company want to understand customer needs/sentiments and adapt quickly to keep their customers happy.

The company also want to leverage their existing IT investments (e.g., APIs/services that they already have, data and knowledge base) to deliver cutting edge GenAI capabilities in **cost efficient manner**.

Live Demo (wait for 30 seconds to restart server on your first request): https://genai-demo.onrender.com/site/ (deployed on render.com for demo)

What it does (Business impact)?

As part of this implementation, a fully functional AI Chat Agent is developed using Google's AI services for a rental car business. The implementation will help both sides (rental car customer as well as rental car business).

What a Rental car business will get?

  • GenAI capabilities will uncover hidden opportunities for a rental car business by learning from existing knowledgebase
  • Allow businesses to cross-sell and upsell services/products by understanding customer needs and sentiments
  • Retain customer by addressing their concerns before they leave and increase brand loyalty
  • Leverage existing IT investments (APIs, services, content, data etc.) to enable GenAI capabilities at lightning speed
  • Easy, Fast and Cost-effective integration of AI/ML and GenAI capabilities to generate more value for business
  • Harness the power AI/ML and GenAI to deliver the best customer experience and exceed the customer satisfaction

What Customers will get?

  • An exceptional experience of interacting with GenAI based chat agent (not like traditional chat bots)
  • Complete tasks like rental car booking, cancellation, booking lookup and provide feedback about service
  • A real human like experience through the interaction with AI Bot
  • Get more contextual information that sometime humans may not have or remember
  • Avail rental car related service 24x7 with same quality and accuracy

How we built it?

Following tools and technologies are used to building this solution/demo:

Google Project ID: woven-mesh-421014 (name: Google AI Hack 2024)

  • Google AI Studio / Gemini ( gemini-1.0-pro-001 - [https://aistudio.google.com/app/prompts])
    • Used for finetuning AI Agent response using prompt engineering/system instructions and example scenarios (few-shot prompting)
    • Used for sentiment analysis of user feedback/comments (using Google Node.js SDK)
    • Used for suggesting an promotional offer to a customer based on sentiment analysis of feedback (if feedback was negative)
    • Read more about implementation
  • Google Function Calling to invoke business APIs/Services
  • AI Chat Assistant There are 2 versions of AI Chat Agent. First one using Low/No-Code Google Vertex AI Agent and, second Custom Chat widget using HTML and JS. The. The Vertex AI Agent will not work for all users while app is in Testing environment.
  • Google - Node.js SDK (@google/generative-ai and @langchain/google-genai)
    • Interact with Gemini via Node.js for building REST APIs
  • NestJS
  • Static Web Page (HTML/CSS)

This is a fully functional AI Chat Agent/Assistant powered by Google AI for a rental car business. The application uses Google AI Studio/Gemini and Vertex AI Agent/Conversation primarily.

The Google AI (Gemini) and Vertex AI Agent capabilities are leveraged to provide an exceptional customer experience to a rental car company. The AI Agent will help customers finding a right car, booking a car, sending email confirmation, cancelling booking, perform sentiment analysis of customer feedback (about rental car experience). If the feedback is negative/unsatisfied, suggest an offer to retain and make them happy.

What is included in the implementation?

  • 1. Google AI (Gemini) Model and Prompts: Leverage prompt template and prompt engineering to interact with customer, analyze customer feedback to determine sentiment and suggest an offer

  • 2. Google Vertex AI Agent: Vertex AI Assistant for interactive chat. Following features are implemented using Vertex AI Assistant:

    • Assist user finding a rental car (with specific make, model, in a particular city etc.)
    • Assist user booking a rental car
    • Use Vertex AI Assistant to collect customer feedback about the rental car service
    • Analyze the feedback (sentiment analysis) dynamically using the Gemini Node.js SDK (integrated via Agent TOOL)
    • Suggest an appropriate offer to customer based on the sentiment analysis of the feedback
    • Use Vertex AI AGENT and TOOL to interact with existing business functions (via existing REST API)
    • Read more about implementation

  • 3. REST API/Open API/Swagger: REST API and Swagger/YAML to create a custom integration (Agent TOOL) for Vertex AI Assistant.

    • REST APIs to perform rental car related operations (list car, booking, cancellaton etc.)
    • Vertex AI Agent will use these API (via TOOL capability of Agent App) to perform business action via chat
    • Vertex AI Agent will be able to provide contextual information/response to customer
    • Gemini integration will analyze customer feedback to determine sentiment and suggest an offer to unsatisfied customers
    • OpenAPI/Swagger definition: https://genai-demo.onrender.com/api
    • Read more about implementation

  • 4. Web Application: A web page/application integrated with Vertex AI Assistant Web Chat for demo.

Challenges we ran into

  • The Google Vertex AI Agent chat can not be used by all users while the App is in Testing phase and hence to demo the live Chat functionality, I ended up creating a HTML and JavaScript widget. I have recorded a separate video to demonstrate Vertex AI Agent in action.
  • A little research and planning was needed to understand how to stitch various GenAI capabilities together
  • Choosing a use case for implementation that demonstrates Google's AI capabilities as well as provides business value was not easy
  • Initially the API integration (via Node.js) was not available for Gemini 1.5 Pro
  • Hosting demo (eventually hosted on a free cloud service render.com)

Accomplishments that we're proud of

  • Able to build and deliver an end-to-end working business use case leveraging Google's capabilities
  • Learning and gaining knowledge about Google's AI capabilities

What we learned

  • This is my first project using Google's AI services and I learned a lot
  • Learned about various capabilities and services offered by Google to enable GenAI capabilities of an individual and businesses
  • Was able to compare Google's services with other similar services

What's next for Customer Experience AI (CxAI)

For this demo, only a small subset of use cases was developed to demonstrate Art of the Possible with Google's AI capabilities. The possibilities with GenAI in customer experience space are exciting and endless. It will be and interesting project to explore further by adding more capabilities

Live Demo

Example Customer Interactions to try with the Vertex AI Assistant

  • Book a car: I want to book a Ford in Toronto, I want to rent a car for 3 days in Toronto
  • Find reservation: find my booking, find my reservation detail
  • Cancel booking: cancel my rental booking
  • Sentiment Analysis: It took us almost three hours just to get a car! It was absurd., The price of car was too high and the whole process took a lot of time.

Want to learn more about the solution?

More details about tools/services that are used for developing the solution.

Google AI Studio

Fine-tune LLM using various prompts to train and optimize model to perform following tasks:

  • Interact (like real human) with customers and assist them with a rental car booking, cancellation etc.

  • Perform sentiment analysis of the feedback provided by customers

  • Based on training (prompts, instructions), suggest suitable offer to an unsatisfied customer

Google Vertex AI Assistant

Below are Visualization of Vertex AI Assistant AGENT and TOOL

  • AGENTS - handle customer interaction

  • TOOLS - invoke external systems/APIs via OpenAPI Integration

NestJS REST APIs and Swagger/Open API

  • Google Gemini Service (Swagger/OpenAPI) - for human like chat using LLMs/Gemini and Sentiment Analysis

  • Rental Car Services - for rental car related service (book a car, cancel booking, search booking)

Image Gallery

  • Google Vertex AI Assistant web page integration integrated with web page: https://genai-demo.onrender.com/site/ Vertex AI Assistant

  • Google Vertex AI Assistant Chat Screens

    • Use case 1: Booking a rental car

      Booking confirmation email:

    • Use case 4: Sentiment analysis and offer suggestion

    • Use case 3: locate/search an existing booking

    • Use case 4: Sentiment analysis and offer suggestion

Thank you!

Built With

Share this project:

Updates

posted an update —

Note about video presentation

The solution was initially implemented used Google Vertex AI Agent but, I realized that I can not get the Chat App approved from Google just for the demo. The Google approval process need formal privacy policy etc. to successfully complete the approval check. Hence, I have changed the implemented to a JS based chat widget instead that Hackathon evaluators can access. I have also recorded a demo using Google Vertex AI Agent App (using my test account). I have shared both videos below.

More detail about the approval process: https://support.google.com/cloud/answer/13464323/#exemptions

Video Demo/Pitch:

Log in or sign up for Devpost to join the conversation.