Our project: VzConnectBot

The Future of Verizon Online Assistance" is a groundbreaking initiative that harnesses the power of conversational Generative AI to revolutionize the digital experience within Verizon's ecosystem. Our vision is to build a chatbot that redefines user interactions by offering personalized and simplified navigation by giving personalized offerings for user queries.

Inspiration:

Our inspiration for this project stems from the immense potential of GenAI and conversational AI to reshape how users engage with Verizon's digital platforms. We are driven by the idea that AI-driven chatbots can provide a seamless and tailored experience, making it easier for customers to explore products, access services, and engage in educational experiences. When we go through the large organizations websites like Verizon and UF which has different departments, provides different services and offerings, We had to go through different links and services to reach our destination web page. This deviates user from their motive and reduces the user experience as they have to go through such a time consuming process and hassle to find that data or service they needed. So, we worked on developing the VzConnectBot, This innovation of ours stems from the belief that the future of Chatbots lies in simplifying the user journey, reducing clicks, and ensuring intuitive navigation and providing conversational real world experience we get with live agents. Our inspiration comes from the desire to create inventive products and plans designed to meet the evolving needs of future customers, ultimately making their digital interactions very effortless, user friendly and highly efficient.

What it does:

VzConnectBot is a cutting-edge customer service chatbot designed exclusively for Verizon customers. Leveraging the power of Generative AI, specifically Open AI(ChatGPT 3.5) fine tuning, it offers personalized and human-like interactions to assist users in real-time. The chatbot's frontend provides an intuitive user interface using ReactJS, while its backend, built using Django, runs a finetuned Open AI( ChatGPT) based generative AI model. This model is trained with the latest data from Verizon's website, ensuring it delivers up-to-date information on products, services, and more. VzConnectBot simplifies customer inquiries, offers accurate responses, and significantly enhances the overall digital experience by providing the right answers for the user queries, support and FAQ. We have trained the the LLM by fine tuning it with the verizon website specific data, which results in our Chat Bot well versed on all the data it is trained upon. So, our VzConnectBot can answer complex or very minute details present in the website with in a matter of few seconds, helping the customers to escape the hassle of going through many services, components or routes to reach their required information. We have also trained our model upon the different routes in the VERIZON.COM. So, when a user requests an info or a service, The bot can respond with both the required information and the specific link to go to, for finding any additional information. We have also added different parameters for our chatbot like Conciseness, Professionalism and Friendliness. These parameters can be adjusted by the user to fetch the output in their required forms and needs.

How we built it:

We built VzConnectBot through a multi-faceted approach:

Frontend: We designed a user-friendly frontend interface to facilitate seamless user interactions using Recat.js, ChakraUI.

Backend: Using Django, we created a robust backend that acts as the backbone of the chatbot, managing user requests and data flow.

Data: The dataset we used for training our ML model is the Verizon website data. We have obtained this data by going through all the pages in Verizon.com and extracting the current data present in those respective web pages. Even though, We have started with having absolutely no idea on our hands , later on we had a lot of data to process on our hands. Even though we took a tad bit more time than we expected on data mining, It is great to see that large data tuned our model well to get great responses.

Generative AI: We trained a state-of-the-art Generative AI model OpenAI (ChatGPT) using Python in Google Collab and Saved the models for using them in our backend. We have trained the model on our dataset of different routes in Verizon.com along with the data available in each website. This model learns from updated data on the Verizon website to provide real-time and accurate information. Furthermore, the entire application is Dockerized for streamlined deployment and management.

Challenges we ran into:

Developing VzConnectBot came with its share of challenges:

Data Integration: Synchronizing data from the Verizon website in real-time proved to be a technical hurdle. Scraping, processing the data, removing all the unnecessary noise from the data took a lot of effort than we anticipated.

User Experience: Striking a balance between human-like interactions and efficiency was a complex challenge. Like when we strictly want the GenAI( Open AI) to provide content from trained veizon data, the regular conversational interactions are affected.

Model Training: Training the Generative AI model with constantly evolving data required ongoing efforts and optimization to find the right tuning. Training on this large Verizon data for fine tuning took a lot of time for model training and data retrieval process. Additionally, ensuring that the dockerized application deployed seamlessly in this limited time has presented its fair set of challenges.

Accomplishments that we're proud of:

Human-Like Service: Achieving a chatbot that delivers near-human customer service interactions on our customer specific non-public data is a significant accomplishment. We planned on using BARD as this GenAI had some more features to offer as it is trained well on current data, However when we came to know that there is no Bard API currently available for use, we have opted for Open AI, which also achieved really good results when we fine tune with our data.

Up-to-Date Information: Ensuring that our AI model stays current with the latest data on the Verizon website is a notable achievement. Moreover, successfully dockerizing and trying to deploy the application in this constricted time period contributes to our sense of accomplishment.

Enhanced User Experience: Providing Verizon customers with a more efficient and personalized digital experience is a source of pride using the state of the art Generational AI model.

Throughout this project, we gained valuable insights:

*The potential of Generative AI in customer service and user experience areas.

*The challenges and rewards of real-time data integration.

*The importance of striking the right balance between automation and human-like interactions.

*Additionally, the experience of dockerizing and deploying the application has added to our knowledge base.

The journey doesn't end here. The future of VzConnectBot involves:

Further AI refinement: Continuous model improvement for even more accurate responses.

Expansion of services: Expanding the chatbot's capabilities to offer a wider range of support other than providing data and few recommendations and specialized web links. We can add more features like authentication, personalization, purchases and payment and security features.

Enhanced user personalization: Implementing advanced user profiling for tailored experiences. Potential integration with voice assistants and mobile apps for a truly seamless customer support experience, all while maintaining efficient application deployment.