-
-
demo picture
Inspiration:
Nikesh has an initiative to launch more robots across our platform and across the company. This will not only improve efficiency, but also reduces the cost.
What it does
It conducts basic conversation with users to address their support requests. Given user's input, the chatbot will reply with the appropriate reply as well as recommend relevant knowledge base articles to the user's query based on the historical users' support ticket content and support engineers' replies.
How I built it
I extracted the training corpus from historical ticket database first. Then I developed a novel machine learning recommendation model on top of open sourced natural language processing framework released by GOOGLE in Nov 2018. Last, I developed a graphical user interface(GUI) to provide a better user experience.
Challenges I ran into
I consider myself a data scientist. I am an expert in machine learning and data science, but I have not done much in frontend coding and has no experience in GUI development before.
Also, the machine learning is over 400M in size due to its 1.2 billion parameters and the corpus is also large in size, which exceeds github submission size cap(100M). I have to host these non-code content in google drive for download, only PANW accounts can view them.
Accomplishments that I'm proud of
I think my chatbot is very useful and has the potential to be applied to a wide range of areas. It is ready to be used, easy to be integrated into current business logic and can be trained with different domain knowledge to adapt to different use cases. Recommending only top 2 related links from the over 130,000 knowledge articles and tech-docs is very challenging. However, the machine learning model developed yields a very high accuracy in offline evaluation. For thousands of custom service tickets since 2018, about 1/3 of total support engineer solved tickets have exactly same links recommended by the chatbot. Prisma cloud Customer success team already expressed strong interests to deploy it to the production.
Log in or sign up for Devpost to join the conversation.