Annex A

LARGE LANGUAGE MODEL CHATBOT SYSTEM USING OPENAI API AND LANGCHAIN
FRAMEWORK

Diesto, Francis Ryan

I. INTRODUCTION

The rapid advancements in artificial intelligence (AI) and natural language processing (NLP)
have paved the way for the development of intelligent conversational agents known as chatbots. These
chatbots have emerged as effective tools for providing automated assistance and enhancing user
experiences across various domains. A chatbot developed by OpenAI called ChatGPT [1], recently
imploded with international popularity because of its ability to generate human-like responses to
questions from various fields of knowledge in any language that is included in its dataset. ChatGPT is a
large language model utilizing a NLP deep learning model called GPT-3, trained on a large dataset of
conversations. ChatGPT’s automation of conversational tasks have significantly saved time and resources
of people, especially work-related tasks such as answering customer queries, personalized customer
conversations, reducing costs in businesses and freeing up resources [1]. ChatGPT is also able to learn
and improve from the prompts it receives and the developing landscape of available data that it can
integrate, making manual updates obsolete.
The usefulness of chatbots in different fields of study are limitless. Students from the Polytechnic
University of the Philippines developed a chatbot called Pharmabot [2]. The chatbot operates on a stand-
alone computer that answers user questions regarding pediatric generic medicine consultation to assist
users in taking the right generic medicine for a certain ailment. The study showed a survey that conducted
the effectiveness of the chatbot. In terms of user-friendliness and the consistency of responses of the
chatbot on the scale of 1 to 5, students identified that the chatbot has a weighted mean of 4.595 and 4.428,
while the medical experts assessed it with a weighted mean of 4.166 and 4.0 respectively. With this, the
acceptability of a developed chatbot proved its usefulness in responding to the needs of users.
The University of Southeastern Philippines (USeP) has launched its own currently work-in-
progress chatbot called SePhi [3]. The chatbot answers questions about general information and inquiries
and is designed to be integrated with websites and other social platforms.

STATEMENT OF THE PROBLEM

This study addresses the need for an efficient and interactive chatbot system powered by the
ChatGPT API within the university environment. Currently, the university experiences challenges in
effectively handling a high volume of inquiries and requests from students, faculty, staff, and other
stakeholders. The existing communication channels, such as email and manual support systems, often
lead to delays in responses and increased workload for university personnel. Additionally, the lack of a
centralized and automated system results in inconsistent and incomplete information being provided to
users.
The following are the research questions that this project aims to answer:
1. How can the implementation of an efficient and interactive chatbot system improve the handling
of inquiries, requests and reduce workload of university personnel?
2. How well does the chatbot system perform, evaluated in terms of response time, accuracy, user
satisfaction, and workload reduction for university personnel?
 3. What are the user perceptions and experiences of utilizing a chatbot system for resolving
inquiries and requests in a university setting?

OBJECTIVE

The objective of the study is to address the need for an efficient and interactive chatbot system
powered by the ChatGPT API within the university environment. The study aims to tackle the challenges
faced by the university in effectively handling a high volume of inquiries and requests from students,
faculty, staff, and other stakeholders. The primary goal is to develop a centralized and automated system
that can improve response times, reduce the workload for university personnel, and provide consistent
and accurate information to users. The study seeks to explore the implementation and impact of this
chatbot system within the university context to enhance communication and streamline support processes.

SCOPE AND LIMITATIONS

This study focuses on developing and evaluating a University Chatbot using ChatGPT API and
Langchain. The research will involve accessing and examining information that is openly accessible and
does not require any proprietary or restricted data. This study also includes gathering feedback and
insights from users, including students, faculty, and staff to assess their satisfaction and experience with
the chatbot system.

DEFINITION OF TERMS

GPT - Generative Pre-trained Transformer. GPTs are language models that have been trained to
comprehend natural language and programming code. They generate text outputs based on the inputs they
receive, which are commonly known as "prompts". Crafting a prompt is akin to instructing or guiding a
GPT model, typically by offering examples or instructions to help it successfully accomplish a given task.
API - Application Programming Interface. API acts as a bridge or intermediary between different
software systems, enabling them to seamlessly communicate and share data. It provides a standardized
way for developers to access and use the functionalities and services of another software application
without needing to understand its internal workings.
LLM - Large Language Model. LLMs are artificial intelligence systems designed to generate
human-like text based on the input it receives. It is trained on vast amounts of textual data to learn
patterns, grammar, and semantics in order to generate coherent and contextually relevant responses.

II. REVIEW OF RELATED LITERATURE

Khin and Soe [4] developed a chatbot system using AIML, an XML-based language for creating
chatbots. Their study aimed to implement an interactive chatbot in Myanmar language at the University
of Computer Studies. The chatbot's knowledge base included 970 question-answer pairs in Myanmar
language. The system utilized the Pandorabots Server and API for hosting and natural language
processing. However, the system had limitations in response time and experienced unexpected responses
due to normalization errors. The researchers proposed solutions such as using substitution files and
expanding the knowledge base for more human-like responses.
Mageira et. al [5] stated that the traditional mentor-centered technique in which students are
always working to stay up, store data, and adjust to the quick pace of the modern instructional period now
reigns in schools in developed nations. However, such a technique is unable to tackle the problems of the
computerized era, which demands exceptional information and skills from understudies and mentors in
order to take advantage of the capabilities offered by Data and Correspondence Advancements (ICT). In
regards to AI that responds to the user in their daily lives, This technology's first generation was centered
on brief task-oriented communication [6]. AI chatbots can serve as intelligent instructors in the
educational setting by delivering instructional information, generating discourse, offering feedback to
students, and so on. In certain circumstances, AI chatbots can supplement or support human instructors by
answering students' queries and offering instruction around the clock, which is obviously difficult or
impractical to achieve with human tutors.

III. METHODOLOGY

Figure 1. Conceptual Framework (LangChain and OpenAI)

RESEARCH DESIGN

This research will be using a quantitative research approach. Quantitative research involves the
collection and analysis of numerical data to measure and quantify various aspects of the chatbot system's
performance. [7] The study will employ surveys as the primary data collection method, allowing for the
systematic gathering of data on response time, accuracy, user satisfaction, and workload reduction.
Surveys will be administered to users, including students, faculty, staff, and site visitors to gather their
perceptions and feedback regarding the chatbot system.

DATA COLLECTION

The data collection in this study will involve a primary data collection in the University of
Southeastern Philippines (USeP) pertaining to documentary data of predetermined responses to frequently
asked questions (FAQ), requirements of requests, and general information about the university services
and the clubs inside the university. Moreover, the researchers propose to relay an evaluation survey form
regarding the user-friendliness, consistency of responses, and the speed of responses for the evaluation of
the chatbot’s effectiveness. The researchers will ask permission to each service inside the university that
caters to the public, especially, students, faculty members, and to those interested in visiting the
university.

RESEARCH METHOD

This study will utilize several methods to develop a website for a USeP chatbot that provides
automated responses to the public's questions about the university, the university services, and the clubs
within the university. The primary methods that this study will be using are as follows:

1. ChatGPT API
ChatGPT is a generative AI chatbot developed by OpenAI that generates responses
according to user prompts [8]. The GPT stands for Generative Pre-trained Transformer referring
to how ChatGPT performs processes and formulate responses in the background. Along with
reinforcement learning through human feedback and reward models, ChatGPT generates human-
like responses by ranking the best responses it can generate according to the user prompt.
OpenAI has an API that provides access to OpenAI’s language models, including GPT-3
and GPT-4 which are models used in ChatGPT. The API has affordable pricing but has usage
limitations which differ for each model. Each model is limited to a finite number of requests and
tokens to call the API that a user can make per minute.
The researchers intend to use the GPT model “text-davinci-003” to be incorporated as the
backbone model for the proposed chatbot system because of its high limit for requests-per-minute
and tokens-per-minute.

2. LangChain
LangChain is a framework for developing language model powered applications that are
able to connect to other sources of data other than the knowledge base of the language model and
allow the model to interact with its environment [9]. The main function of LangChain is to
generate components with multiple implementations for working with language models which are
modular and easy-to-use, and the off-the-shelf chains which acts as an assembly of components to
accomplish higher level tasks like constructing a sequence of calls to the API or the language
model. LangChain is mainly used for creating chatbots, answering questions using custom data
sources, and analyzing structured data.
The researchers will utilize LangChain to integrate custom data sources, specifically PDF
files of the documentation of university services for the GPT model to generate responses from.
This will be according to the university services’ documentation of requirements and general
information.

3. Flask
Flask is a backend web framework that is used to develop web applications using Python
scripts [10]. Flask is lightweight, easy-to-use and its implementation is straightforward. Flask's
simplicity and modular nature make it a popular choice for building small to medium-sized web
applications and APIs. Its lightweight design allows developers to have more control over their
projects and choose the components they need.
The researchers intend to use Flask as the main rendering framework for the template and
function of the web application with Python scripts to be integrated to websites such as the USeP
main landing page, and the USeP Portal.

4. JavaScript
JavaScript is a lightweight, interpreted, and compiler based scripting programming
language with first-class functions that makes web applications dynamic and interactive for users
 [11]. JavaScript's versatility and widespread adoption have made it a crucial language for web
development. It allows developers to create interactive user interfaces, build web applications
with complex functionality, and provide a seamless user experience. The researchers will utilize
JavaScript as the main programming language to add interactivity and dynamical aspects to the
web application.

ETHICAL CONSIDERATIONS

The study will adhere to ethical standards and follow the ethics board protocol of the University
of Southeastern Philippines. The researchers will ensure voluntary participation, allowing participants to
join the study without facing any negative consequences, and they will respect their decision to withdraw
from the study, deleting any collected data. The privacy and confidentiality of participants will be upheld,
with their names kept confidential through the use of aliases. The consent form used will be easily
understandable, providing a clear perspective on the potential benefits of the study. Participants will
receive informed consent, informing them of the risks, benefits, disadvantages, and advantages associated
with participating.

REFERENCES

[1] J. Deng and Y. Lin, “The benefits and challenges of chatgpt: An overview,” Frontiers in Computing
and Intelligent Systems, vol. 2, no. 2, pp. 81–83, 2023. doi:10.54097/fcis.v2i2.4465

[2] B. E. Comendador, B. M. Francisco, J. S. Medenilla, S. M. Nacion, and T. B. Serac, “Pharmabot: A

[3] USeP PAD, “USEP deploys chatbot AI ‘SePhi’ to enhance university experience,” University of
Southeastern Philippines, https://www.usep.edu.ph/blog/2023/05/02/usep-deploys-chatbot-ai-sephi-
to-enhance-university-experience/#prettyPhoto (accessed Jun. 26, 2023).

[4] N. N. Khin and K. M. Soe, "University Chatbot using Artificial Intelligence Markup Language," 2020
IEEE Conference on Computer Applications(ICCA), Yangon, Myanmar, 2020, pp. 1-5, doi:
10.1109/ICCA49400.2020.9022814.

[5] K. Mageira et al., “Educational AI chatbots for content and Language Integrated Learning,” Applied
Sciences, vol. 12, no. 7, p. 3239, 2022. doi:10.3390/app12073239

[6] Ram et al., “Conversational AI: The Science Behind the Alexa Prize” Alexa Prize Proceedings, 2017,
doi: https://doi.org/10.48550/arXiv.1801.03604.

[7] J. Kamiri and G. Mariga, “Research methods in machine learning: A content analysis,” International
Journal of Computer and Information Technology(2279-0764), vol. 10, no. 2, 2021.
doi:10.24203/ijcit.v10i2.79

[8] A. Hetler, “What is ChatGPT? everything you need to know,” WhatIs.com,