It took over six months of hard work and determination for invenioLSI employee Narayana Rallabandi, Senior Principal Architect, with the support of former employee Akash Singh and a dedicated research intern Srikari Rallabandi, to write a comprehensive research paper titled Webservices Learning Bot: A Responsive, Actionable AIML BOT Framework. The paper was published in the International Journal of Innovative Research in Science, Engineering and Technology with a high impact factor* of 8.118.
The paper is based on a unique bot code that was at the center of four invenioLSI submissions during the MuleSoft Hackathon in 2020 – an annual event where entrants submit innovative MuleSoft projects. The bot code's framework adds dynamism to essentially static conversational bots (chatbots) that are applications used to provide a human-like response experience. Often, these AI-enabled digital assistants only churn out pre-fed information and are unresponsive to queries outside their fields. In this paper, the team presented a practical solution to developing a responsive, actionable, and dynamic chatbot that can interpret the ‘service’ required and provide optimum results in the field computation, using service-generated metadata as the response.
We talked to Narayana and his team to better understand who they are, what inspired them to write this research paper, the journey that followed, the challenges they faced, and the relevance of the paper for the tech community.
1. Tell us a little about yourself
Narayana: At invenioLSI, I work across multiple projects in various areas, from pre-sales to deployment in client environments and building internal product toolsets, helping invenioLSI to deliver client responsibilities. I work on multiple technologies ranging from Java/Java EE, MuleSoft, and Spring, to Python and Django. Previously, I’ve held the role of a Consulting Member, Technical Staff at Oracle, and Staff Engineer at Sun Microsystems. Essentially, I've been building products in the enterprise integration and cloud for a few decades.
Srikari: I'm a third-year B. Tech student majoring in Artificial Intelligence (AI) at VJIT Hyderabad and interned at invenioLSI until December 2022. I am a problem solver, pattern finder, and researcher at heart. Math and statistics have always been a passion of mine. The field of machine learning (ML) and AI is my calling to the research wing of the tech field.
2. What motivated you to write this paper, and how did you come up with this topic?
This paper was a direct result of the MuleSoft Hackathon in 2020. While writing the bot code for the project, we realized the idea was unique in that it beautifully combines the concept of ML and AI to generate a user-specific bot. So, we decided to propose the framework for it in the form of a research paper to enrich the community as well.
3. Can you offer some insight into the research paper?
This paper discussed the possibility of an ML-based Alicebot (as an extension to Program AB), which can learn over the RAML, WSDL, Open API, or Rsyslog. The bot can also be adaptable at a commercial level. It can invoke ‘pluggable’ actions for messaging or escalate-based workflows that can be responsive and helpful in decision-making. The solution we discussed is a framework that is extendable and adoptable with learning over other standards, apart from the ones mentioned. The Actions Framework can plug in any custom action as per the need of the application.
4. How is the bot you have created different from the currently available conversational bots?
Richard Wallace, in 1995, developed a chatbot called Alice, a natural language processing (NLP) chatbot designed to engage in a conversation by reacting to human input and responding as naturally as possible. However, being a basic chatbot, Alice does not serve a purpose as it cannot serve a customer. In 2013, Google released Program AB, an implementation of AIML 2.0, the language on which Alice was built. Our chatbot framework extends this functionality, using Program AB to serve a customer, giving them user-specific outputs. The administrator can monitor and maintain the systems using chat-based commands, so the user doesn't need to be at his desk all the time. The current bots lack this feature.
5. What makes this bot you developed particularly special?
Our bot has been developed using an open source which does not require any costs. This intellectual bot is generic and can be adapted by most web services like CloudHub and SAP services. It is also extendable over various data protocols like databases and CRM, which can expose APIs. Its serverless deployment makes it low-cost and requires less maintenance, offering greater scalability and flexibility. The bot is ‘plug and play’ where you feed it a set of APIs, and it dynamically generates AIMLs for the end user.
6. How long did it take to research and write this paper, and what was the methodology?
Writing the base code took around two months and converting this into a research paper took four months. However, the literature review and finding a suitable journal to publish the work took up most of the time. All this work (writing the research paper) was a side project on top of usual company-related work, so we had to devote 2-3 hours daily to complete it on time, over our regular working hours. We had to confirm the ingenuity of the idea, for which we went through 30-odd research papers. After writing the paper, we had to find a publication suitable for the report, which meant making formatting modifications to conform to the publication's needs and then submitting it for acceptance.
7. How can your paper help the tech community?
This paper proposes a framework that is flexible, extendable, and adaptable. The framework can be implemented across any bot ecosystem, not just Program AB, to bring dynamism into the response system of bots. It can be extended with NLP and other ML modules that can bring context to the web services they are working with. The action API can either use an existing action like notification onto messaging systems like Telegram, or extend to custom alerts and actions with newer protocols and APIs with suggested alerts or actions. These events can be fed into the API ecosystem, making the APIs interactive.
8. How challenging was it to get all the data for this paper?
Srikari: I was working on my project WagginTails (a one-stop pet shop), for which I wanted a chat component. While searching for libraries and current research thoughts, I came across invenioLSI's submissions around the chatbots for the MuleSoft Hackathon. I reached out to those who submitted the project so that I could collaborate with them. While comparing the other libraries, I found it more dynamic and agile.
Luckily, I chanced upon Razia, an Associate Consultant, at invenioLSI, who was an integral part of coding the idea during Hackathons and was a guiding force behind this paper. I learned that invenioLSI was trying to write a paper around the implementation, and I volunteered to help. I was tasked with establishing the bot's uniqueness by doing research and a literature review. The work was not easy as I had to sift through lots of papers (about 70), get the validation about the relevance, and then go back. Finally, I could reduce the number to around 18 documents that are relevant and summarize them as part of the background work and literature review. It's almost a 10-week journey which helped me understand how we do research, how a research paper is written, and what it takes to publish the article.
9. Why do you think it is essential to delve into and write a paper like this?
There are a couple of reasons for this. One is the process that goes through peer group review and can endorse the nature of your work, leading to the acknowledgment that it is essential to research. This brings into the organization the aspect of scientifically recording work you are proud of, encouraging others to emulate. Second, such work could enhance the company's Intellectual Property (IP) and contribute to the community.
This achievement was later followed by an equally significant one where Akash and Srikari wrote a second research paper based on the work done in the computational area of the DevOps dashboard. The paper was recently published in the International Journal For Innovative Engineering and Management Research – a peer-reviewed open-access international journal. The paper delves into various log processors in the area and their flaws. It essentially means that it shouldn't matter where the log originates if it has some common features like log level and timestamps error description in each line. The log parser will pre-process and cluster them in specific relevant groups. In the paper, Akash and Srikari have proposed a model for the new processor which would have better efficiency than the existing ones.
invenioLSI is incredibly proud of Narayana and his team's out-of-the-box thinking and dedication, which resulted in these impressive achievements. The publication of their papers is a significant matter of prestige, and we plan to leverage it to improve and streamline our offerings to clients and within the organization. As part of this effort, invenioLSI is currently developing a DevOps dashboard and plans to integrate the chatbot into it.
*Impact factor is a metric that measures the number of times an average paper in a journal is cited during a year, which indicates the importance of a journal to its scholarly field.