Towards a RAG-Based WhatsApp Chatbot for Animal Certification
Platform Support
Gabriel Vieira Casanova
a
, Pedro Bilar Montero
b
, Alencar Machado
c
and Vinicius Maran
d
Laboratory of Ubiquitous, Mobile and Applied Computing (LUMAC), Federal University of Santa Maria,
Santa Maria, Brazil
Keywords:
Question Answering Systems, Retrieval-Augmented Generation (RAG), WhatsApp, Virtual Assistant,
Automated Support, Chatbot, PDSA-RS.
Abstract:
Ensuring compliance with animal production certification requirements is often a complex and time-
consuming task. This paper presents a domain-specific chatbot designed to assist users in requesting certi-
fications within the PDSA-RS framework. By leveraging Retrieval-Augmented Generation (RAG) and large
language models (LLMs), the proposed system retrieves relevant information from specialized documents and
generates accurate, context-driven responses to user queries. The chatbot’s performance was evaluated on two
Brazilian certification platforms, demonstrating its potential to streamline certification requests, reduce errors,
and enhance user experience.
1 INTRODUCTION
Chatbots have become an integral part of modern
communication, assisting users in tasks such as cus-
tomer support, information retrieval, and service au-
tomation. However, traditional chatbots often face
significant limitations, including a lack of flexibility,
inability to handle complex queries, and reliance on
predefined scripts or rule-based logic. These short-
comings make it challenging for them to provide ac-
curate, context-aware responses in dynamic scenarios
(Kucherbaev et al., 2017).
The advent of Large Language Models (LLMs)
has addressed many of these challenges by en-
abling chatbots to understand and generate human-
like responses with remarkable fluency and contextual
awareness. Yet, despite their capabilities, LLMs have
their own limitations. They can struggle with retriev-
ing accurate, up-to-date information, as their training
data is static and finite, and may generate plausible
but incorrect answers when faced with queries beyond
their knowledge (Zhao et al., 2023).
To overcome these limitations, Retrieval-
Augmented Generation (RAG) offers a transforma-
a
https://orcid.org/0009-0009-5420-7334
b
https://orcid.org/0009-0002-9224-7694
c
https://orcid.org/0000-0003-2462-7353
d
https://orcid.org/0000-0003-1916-8893
tive solution. By integrating a retrieval mechanism
into LLMs, RAG enables chatbots to access real-
time, relevant information from external knowledge
sources or documents. This synergy between retrieval
and generation allows chatbots to move beyond pre-
trained data, delivering more accurate responses,
context-aware, and dependable (Zhao et al., 2024).
The Plataforma de Defesa Sanit
´
aria Animal do
Rio Grande do Sul (PDSA-RS) is a control system
designed to support animal health management by or-
ganizing and integrating all stages of certification pro-
cesses for poultry and swine farming in Rio Grande
do Sul, Brazil. While the platform is a critical tool
for ensuring operational efficiency and compliance,
users often encounter frequent questions and chal-
lenges when navigating its functionalities. These re-
curring user questions highlight the need for a sup-
port system capable of providing accurate and imme-
diate assistance regarding the platform’s use (Descovi
et al., 2021). WhatsApp, as one of the most widely
used messaging platforms globally, offers an ideal
medium for deploying such advanced chatbots. By
incorporating a RAG-based chatbot into WhatsApp,
users of the PDSA-RS platform can access intelli-
gent, dynamic assistance tailored to their needs. The
chatbot is specifically designed to address common
user questions about the platform, providing accurate
and context-aware responses to streamline user expe-
rience and reduce support overhead (Times, 2021).
Casanova, G. V., Montero, P. B., Machado, A. and Maran, V.
Towards a RAG-Based WhatsApp Chatbot for Animal Certification Platform Support.
DOI: 10.5220/0013473500003929
Paper published under CC license (CC BY-NC-ND 4.0)
In Proceedings of the 27th International Conference on Enterprise Information Systems (ICEIS 2025) - Volume 1, pages 1077-1084
ISBN: 978-989-758-749-8; ISSN: 2184-4992
Proceedings Copyright © 2025 by SCITEPRESS – Science and Technology Publications, Lda.
1077

This paper presents the development of a RAG-based
Chatbot for WhatsApp, designed to support users of
the PDSA-RS platform. Leveraging the combined
strengths of LLMs and retrieval systems, the chat-
bot aims to deliver accurate, contextually aware in-
teractions, enhancing user engagement and improving
understanding of the platform’s functionalities (Gao
et al., 2018).
The paper is structured as follows: Section 2 pro-
vides an overview of the key background concepts rel-
evant to the research and the proposed approach. Sec-
tion 3 introduces the methodology for developing the
Chatbot for the Rio Grande do Sul Animal Health De-
fense Platform (PDSA-RS), specifically designed to
provide support through WhatsApp. The simulation
of the proposed solution is detailed in Section 4. Fi-
nally, Section 5 presents the conclusions of this work.
2 BACKGROUND
2.1 Chatbots
Chatbots are software applications engineered to
mimic human conversation, predominantly through
text-based interfaces. Their development has seen
substantial advancements, transitioning from basic
rule-based systems to sophisticated AI-powered con-
versational agents (Pantano and Pizzi, 2023).
The inception of chatbots dates back to the mid-
20th century. A notable milestone was achieved in
1966 with the creation of ELIZA by Joseph Weizen-
baum. ELIZA utilized pattern matching and substitu-
tion techniques to simulate dialogue, setting a founda-
tional precedent for subsequent advancements in nat-
ural language processing (NLP) and conversational
AI (Weizenbaum, 1966).
The 2010s witnessed a pivotal transition towards
AI-driven chatbots. The emergence of machine learn-
ing and NLP technologies significantly enhanced
chatbots’ ability to comprehend and generate human-
like responses. Platforms such as IBM Watson, Mi-
crosoft Bot Framework, and Google’s Dialogflow
provided developers with robust tools to create more
intelligent and versatile chatbots (Wolf et al., 2019).
The 2020s have brought further innovations in
conversational AI, with chatbots becoming increas-
ingly context-aware and capable of managing intri-
cate queries. Large Language Models (LLMs) have
been instrumental in this progress, enabling chatbots
to understand and produce text in a manner akin to
human communication. Companies are progressively
integrating chatbots into diverse applications, ranging
from customer service to personal assistants and edu-
cational tools (Radford et al., 2019).
The evolution of chatbots, from simple rule-based
systems to advanced AI-driven softwares, has been
marked by continuous innovation. Driven by ad-
vancements in NLP and AI, chatbots have become in-
tegral to modern technology. As research progresses,
chatbots are set to further transform user experiences
across various domains (Adamopoulou and Moussi-
ades, 2020).
2.2 WhatsApp
WhatsApp is a popular messaging app that allows
users to send text messages, make voice and video
calls, and share media files. It was founded in 2009
by Brian Acton and Jan Koum and was acquired by
Facebook in 2014. WhatsApp has become one of the
most widely used communication tools globally, with
over 2 billion users.
WhatsApp offers a variety of features that make
it a versatile communication tool, including text mes-
saging, voice and video calls, media sharing, status
updates, and end-to-end encryption. Its widespread
adoption and continuous updates make it a reliable
choice for both personal and professional communi-
cation (Times, 2021).
2.3 Large Language Models
Large Language Models (LLMs) represent a signifi-
cant advancement in the field of artificial intelligence,
particularly in natural language processing. These
models have evolved rapidly, driven by innovations
in deep learning and the availability of vast amounts
of text data (Chang et al., 2024).
The 2010s saw the introduction of transformative
models like Word2Vec and GloVe, which focused on
word embeddings. The real breakthrough came with
the introduction of the Transformer architecture by
Vaswani et al. in their 2017 paper ”Attention is All
You Need.” This architecture, based on self-attention
mechanisms, allowed for more efficient processing of
sequential data, paving the way for larger and more
complex language models (Singh, 2024).
In 2018, Google introduced BERT (Bidirectional
Encoder Representations from Transformers), which
revolutionized the field by demonstrating state-of-
the-art performance on a wide range of natural lan-
guage processing tasks. The early 2020s witnessed a
proliferation of LLMs, with models like RoBERTa,
T5 (Text-to-Text Transfer Transformer), and others
building on BERT’s success. Notably, the introduc-
tion of models like ChatGPT by OpenAI showcased
ICEIS 2025 - 27th International Conference on Enterprise Information Systems
1078

the practical applications of LLMs in conversational
AI (DATAVERSITY, 2024).
2.4 Retrieval-Augmented Generation
Retrieval-Augmented Generation (RAG) is a tech-
nique that combines information retrieval (IR) with
Large Language Models (LLMs). This method en-
hances the capabilities of language models by retriev-
ing relevant documents or pieces of information from
an external corpus and using that retrieved informa-
tion to generate more accurate and contextually rele-
vant responses (Gupta et al., 2024).
Traditional text generation models rely heavily
on the data they have been trained on and their
internal parameters to generate outputs. However,
these models can sometimes lack the ability to ac-
cess specific, up-to-date, or domain-specific knowl-
edge. RAG overcomes these limitations by incorpo-
rating an external information retrieval step into the
model pipeline, which consists of two main stages:
• Retrieval Phase. The first step involves retrieving
relevant documents or pieces of text from a large
external corpus, typically using an information re-
trieval system.
• Generation Phase. After retrieving the relevant
information, a generative model is employed to
process the retrieved documents along with the
original input, all in a single prompt, and gener-
ate a final response.
The usage of Retrieval-Augmented Generation
has a set of advantages:
• Access to External Knowledge. RAG leverages
external sources like databases, files, or websites
to enhance responses.
• Up-to-date Information. The retrieval step can
provide the latest information, keeping RAG mod-
els current, especially for rapidly changing topics.
• Domain-Specific Information RAG can be tai-
lored to specific domains, enabling expert-level
generation for niche topics.
• Reduced Model Size. By relying on external re-
trieval, RAG models avoid storing all knowledge
internally, resulting in a more efficient model.
In the context of enterprise information systems,
RAG can be applied to tasks such as automated query
answering, document summarization, and knowledge
management. For example, RAG could be used to
provide real-time, evidence-backed answers to em-
ployee queries, summarize complex legal or technical
documents, or assist in customer service by retrieving
and generating responses from enterprise knowledge
bases (Li et al., 2022).
2.5 PDSA-RS
The Plataforma de Defesa Sanit
´
aria Animal do Rio
Grande do Sul (PDSA-RS) is a web platform that of-
fers tools to official veterinarians in animal sanitary
control in Rio Grande do Sul state (Brazil). Using ad-
vanced technologies, the platform provides real-time
data to support decision-making, with features such
as epidemiological analysis, disease spread modeling,
and the evaluation of control measures. The platform
is continuously evolving to address emerging chal-
lenges in animal health (Perlin et al., 2023).
PDSA-RS can be accessed via a web platform and
a mobile application. The web platform allows for
detailed data analysis and visualization, ideal for pro-
fessionals needing in-depth information on disease
trends and control efforts. The mobile app provides
convenient, on-the-go access, enabling users to report
incidents, view outbreak details, and receive alerts.
This dual-access design enhances communication and
supports quick responses to disease threats, helping to
strengthen animal health defenses in Rio Grande do
Sul.
The platform includes a support center that is
hosted in WhatsApp to help users use its features ef-
fectively. An automated chatbot is being developed to
further streamline support, providing immediate help
with common inquiries and guidance on using the
platform. This feature aims to reduce waiting times
and improve user experience by offering quicker ac-
cess to necessary information and troubleshooting di-
rectly through WhatsApp.
3 METHODOLOGY
This section presents the methodology for developing
the Chatbot for the Plataforma de Defesa Sanit
´
aria
Animal do Rio Grande do Sul (PDSA-RS), specifi-
cally designed to provide support through WhatsApp.
The focus is on establishing a robust architecture that
integrates with the existing platform. The approach
involves a modular design, ensuring scalability and
flexibility in development. The architecture (Figure
1) includes distinct layers responsible for user inter-
action, data processing, and system management. The
development process has followed an agile methodol-
ogy, allowing for continuous improvement and adap-
tation based on feedback and testing throughout the
implementation stages.
Towards a RAG-Based WhatsApp Chatbot for Animal Certification Platform Support
1079

Figure 1: The proposed RAG Architecture.
3.1 Connection Module
This module is responsible for the direct integration
with WhatsApp, ensuring the reception and sending
of messages. When the server starts, a QR code is
displayed in the terminal to pair the connection with
a device. After that, the server listens for messages
from phone numbers, with no restriction to contacts.
All incoming messages are forwarded to the synchro-
nization module for further handling, while all outgo-
ing messages are processed via the processing mod-
ule.
The technologies employed in the system module
are summarized in Table 1. This table provides an
overview of the core technologies used for message
handling, system communication, and WhatsApp in-
tegration, essential to the platform’s functionality.
Table 1: Employed Technologies.
Technology Description
TypeScript A JavaScript superset that adds
static types.
AMQPLib A library for interacting with Rab-
bitMQ in Node.js.
Node.js A runtime for server-side develop-
ment.
NPM A package manager for dependen-
cies.
WPPConnect A library for integrating WhatsApp
with Node.js.
3.2 Synchronization Module
This module addresses the need to reassemble frag-
mented messages, a common behavior among What-
sApp users. By ensuring that long messages split
across multiple parts are properly synchronized, it
maintains message integrity before forwarding to the
processing module.
When a message fragment is received, the mod-
ule checks Redis for an existing entry. If found, it
appends the fragment; if not, it creates a new entry.
The fragments are queued for processing after a 40-
second timer. If a new fragment arrives during this
time, the timer resets to 40 seconds. After 40 seconds
timer times out, we assume the user has finished his
question and the complete message is passed to the
next module.
The technologies employed in this system’s mod-
ule are summarized in Table 2. This table provides an
overview of the core technologies used for message
synchronization.
Table 2: Employed Technologies.
Technology Description
Java A popular programming language
used for server-side applications.
Spring A framework for building Java-
based server-side applications.
Redis An open-source in-memory data
structure store used for caching
3.3 Processing Module
The Processing Module applies the Retrieval-
Augmented Generation (RAG) technique to retrieve
and generate a response based on the question, re-
turning it to WhatsApp. Its operation involves re-
ceiving the defragmented message, i.e., the complete
user question, and performing a similarity search in a
vector database. This query returns similar questions
with their respective answers, enhancing the context
so that the LLM generates a human-like response that
aligns with the user’s intent. When the response is
ready, it is sent to a queue, where the Connection
Module listens and forwards it to the respective user
on WhatsApp. The sequence of steps that defines the
operation of the Processing Module is outlined as fol-
lows:
3.3.1 Document Ingestion
This step involves preparing the knowledge base by
processing a JSON file containing the most relevant
questions and answers from users of PDSA-RS and
converting them into vector representations suitable
for retrieval. The sub-processes include:
• Text Extraction. Extracting specialized query
collected relevant text data to a JSON file by con-
sulting team’s members a from user support at the
help center of PDSA-RS.
ICEIS 2025 - 27th International Conference on Enterprise Information Systems
1080

• Partitioning or Chunking Strategy. Splitting
the text content into chunks of size 7500 tokens
with an overlap of 100 tokens to preserve contex-
tual continuity. This Chuncking technique is sep-
cifically denominated as Sliding Window Chunk-
ing.
• Embedding or Vectorization. Transforming the
text chunks into vector representations using the
nomic-embed-text vectorization model. These
embeddings are then stored in the PostgreSQL
database integrated with PGVector for efficient re-
trieval.
3.3.2 Vector Storage
The vector representations generated during the in-
gestion process are stored in a PGVector-enabled
PostgreSQL database, optimized for similarity
search.
3.3.3 Retrieval
A similarity search is performed on the vector
database using the vectorized representation of the
user’s query to find the most relevant chunks of text
that match the user’s question.
3.3.4 Response Generation
With temperature set to zero ensuring precise and de-
terministic outputs, the retrieved information is used
to enhance the context, enabling a language model,
Ollama with LLaMA 3.2 7B, to generate a coherent
and human-like response aligned with the user’s in-
tent. This configuration minimizes randomness, en-
suring alignment with the setup.
3.3.5 Employed Technologies
The technologies employed in this system’s module
are summarized in Table 3.
Table 3: Employed Technologies.
Component Description
nomic-embed-text Large context length text en-
coder used for vectorizing
text chunks.
LLaMA 3.2b Large language model used
for generating human-like
responses.
Pip Package used to manage de-
pendencies.
Python Programming language.
3.4 Docker Infrastructure
The infrastructure supporting the proposed Chatbot is
built on Docker to ensure modularity, scalability, and
simplified deployment. The tech-stack is summarized
in Table 4.
Table 4: Docker Infrastructure.
Component Description
RabbitMQ Message broker for asynchronous
communication with a monitoring
interface.
Redis In-memory data store for message
fragments, with Redis Insight for
monitoring.
PostgreSQL A relational database for storing
structured data.
PgVector Extension for PostgreSQL enabling
vector similarity search.
PgAdmin Interface for managing PostgreSQL
databases.
Ollama Server A server to get up and running with
large language models.
Open WebUI Web interface for system config-
uration and monitoring of Ollama
Server.
3.5 System Design
The system design of the proposed chatbot provides
a high-level view of its architecture, outlining the
main modules and their interconnections. It illustrates
how these modules collaborate to achieve the chat-
bot’s functionality. The design highlights the flow of
data, interactions between modules, and the integra-
tion points necessary for the system’s operation. This
image serves as a blueprint for understanding the sys-
tem’s overall structure and its cohesive operation (see
Figure 2).
4 SIMULATION
4.1 Objectives
The goal of the simulation is to test the behavior of the
proposed Chatbot in a controlled WhatsApp environ-
ment. The simulation aims to emulate a user-chatbot
interaction by leveraging two smartphones, where one
device acts as the user and the other as the interface
for the chatbot connected to the server.
Towards a RAG-Based WhatsApp Chatbot for Animal Certification Platform Support
1081

Figure 2: The Complete Chatbot System Design.
4.2 Requirementes
The simulation setup involves the following compo-
nents and roles:
• Smartphone 1 (User). Sends messages (ques-
tions or prompts) to the chatbot, emulating a typi-
cal PDSA-RS user interaction.
• Smartphone 2 (Chatbot). Acts as the interface
for the chatbot, paired with a server that processes
incoming messages and generates appropriate re-
sponses.
• Server. Runs the proposed chatbot application,
including components for receiving, processing,
and responding to messages. A computer is re-
quired to run the server, which is responsible for
handling all the message processing tasks.
• Extracted Conversations. From early 2023 to
late 2024, five conversations with PDSA users
were extracted to ingest in the vector database
recurring questions. Each file contained over
500 lines, highlighting common challenges and
needs. This data offers valuable insights to im-
prove PDSA’s support and user experience.
4.3 Configuration
The servers for this simulation are executed on a
Lenovo 3i Laptop with the following specifications:
• Processor. Intel Core i5 10th Gen
• RAM. 8 GB DDR4 3200MHz
• Storage. 256GB SSD
• Operating System. Arch Linux
• GPU. NVIDIA GeForce GTX 1650 with 4 GB of
dedicated memory
4.4 Conditions
To ensure controlled and repeatable testing behavior,
the following conditions are maintained:
• Both smartphones must be connected to a stable
network.
• The server is pre-configured with the chatbot ap-
plication and necessary dependencies.
• A predefined set of test messages (e.g., FAQ-style
questions) is used for consistency in testing.
4.5 Experiment
The experiment scenario is divided as follows:
• Message Transmission. Smartphone 1 sends a
text message to Smartphone 2 via WhatsApp.
The message mimics a typical PDSA-RS user in-
quiry, such as ”O que
´
e a PDSA-RS?” (What
is PDSA-RS?) or ”Como fac¸o para realizar uma
nova solicitac¸
˜
ao de certificado para uma granja de
su
´
ıno?” (How do I submit a new certificate request
for a swine farm?).
• Processing. Smartphone 2, paired with the server,
receives the message. The server processes the
text using the chatbot’s architecture, which in-
cludes natural language understanding, response
generation, and message formatting.
• Response. The connection server awaits the
message from the Processing server and sends
the generated response back to Smartphone 1 on
WhatsApp, completing the interaction.
4.6 Results
The expected outcomes of the simulation include:
• Smartphone 2 successfully receives and forwards
messages to the server.
• The server correctly processes the input and gen-
erates coherent, contextually relevant responses
using the RAG technique.
• Smartphone 1 receives accurate responses in real-
time, effectively mimicking a natural chatbot-user
interaction.
The final and expected output of the real simula-
tion is displayed in Table 5:
5 CONCLUSION
This paper presented the need to automate customer
service at PDSA-RS’s call center and proposed the
ICEIS 2025 - 27th International Conference on Enterprise Information Systems
1082

Table 5: The Simulated PDSA user-chatbot interactions (in
Portuguese).
Smartphone 1 (User
Input)
Smartphone 2 (Chatbot
Response)
Como posso cadas-
trar um novo RT na
PDSA?
Para o cadastro como RT,
voc
ˆ
e pode optar por ser
independente ou repre-
sentante de uma empresa.
Se voc
ˆ
e
´
e um RT inde-
pendente,
´
e necess
´
ario
solicitar o cadastro di-
retamente na plataforma
PESA pelo e-mail registro-
[e-mail].
Se voc
ˆ
e est
´
a vinculado
a uma empresa, o RT
representante da empresa
tem a possibilidade de
cadastrar novos usu
´
arios
diretamente pelo m
´
odulo de
gest
˜
ao. Ap
´
os o cadastro,
o novo usu
´
ario receber
´
a
automaticamente um e-mail
com os detalhes de acesso,
incluindo login e senha.
E Preciso fazer uma
nova solicitac¸
˜
ao de
certificado para uma
granja de su
´
ıno,
como fac¸o?
´
E necess
´
ario que voc
ˆ
e es-
teja como RT da granja em
quest
˜
ao. Ap
´
os,
´
e necess
´
ario
a realizac¸
˜
ao de uma co-
leta, a inserc¸
˜
ao de no min-
imo dois relat
´
orios trimes-
trais e o preenchimento do
formul
´
ario de tuberculina e
leptospirose. Tendo todos
os dados necess
´
arios, basta
acessar o item ”Solicitar
Certificado” no menu lat-
eral da plataforma.
implementation of a chatbot capable of providing
contextualized responses to PDSA, using RAG and
LLM techniques. Many technologies were employed
in each module of the application to compose the to-
talitty of the RAG based WhatsApp Chatbot proposal.
The main contributions include the automation of
customer service, improving efficiency and availabil-
ity of the service, and the application of multiple tech-
nologies to solve real-world problems.
To further enhance the Chatbot and maximize its
impact, it’s recommend expanding its functionalities,
integrating with other platforms, and conducting live
user testing in the real-world scenario of PDSA-RS to
evaluate the long-term effects of the Chatbot imple-
mentation.
• Monitoring Dashboard. Develop a web inter-
face to monitor questions and answers in real
time, enabling performance analysis and the iden-
tification of areas for improvement.
• Conversation History Management. Implement
a module to manage and easily access interaction
history, assisting in analysis and future chatbot
training.
• Scalable Solutions. Replace PGVector with more
robust tools, such as Elasticsearch, for vector
queries.
• Long-Term Impact Evaluation. Conduct stud-
ies to assess the effects of the chatbot on user ex-
perience and organizational efficiency over time.
• Multimodal Agents. Utilize LLMs that process
not only text but also images, documents, and
even audio, expanding the chatbot’s capabilities.
• Real Environment Testing. It is necessary to test
the chatbot in a real test environment, with real
users from PDSA-RS, as the paper only covers a
simulation.
• Evaluation of Response Effectiveness. To en-
sure continuous improvement, it is essential to ap-
ply metrics that assess the relevance, precision,
and overall quality of the chatbot’s responses.
ACKNOWLEDGEMENTS
This research is supported by FUNDESA (project
’Combining Process Mapping and Improvement with
BPM and the Application of Data Analytics in the
Context of Animal Health Defense and Inspection
of Animal Origin Products in the State of RS -
UFSM/060496) and FAPERGS, grant n. 24/2551-
0001401-2. The research by Vin
´
ıcius Maran is par-
tially supported by CNPq, grant 306356/2020-1 -
DT2.
REFERENCES
Adamopoulou, E. and Moussiades, L. (2020). Chat-
bots: History, technology, and applications. Machine
Learning with Applications, 2:100006.
Chang, Y., Wang, X., Wang, J., Wu, Y., Yang, L., Zhu, K.,
Chen, H., Yi, X., Wang, C., Wang, Y., et al. (2024). A
survey on evaluation of large language models.
DATAVERSITY (2024). A brief history of large language
models - dataversity.
Descovi, G., Maran, V., Ebling, D., and Machado, A.
(2021). Towards a blockchain architecture for animal
sanitary control. In ICEIS (1), pages 305–312.
Gao, J., Galley, M., and Li, L. (2018). Neural ap-
proaches to conversational ai: Question answering,
Towards a RAG-Based WhatsApp Chatbot for Animal Certification Platform Support
1083

task-oriented dialogues and social chatbots. Founda-
tions and Trends in Information Retrieval.
Gupta, S., Ranjan, R., and Singh, S. N. (2024). A com-
prehensive survey of retrieval-augmented generation
(rag): Evolution, current landscape and future direc-
tions.
Kucherbaev, P., Psyllidis, A., and Bozzon, A. (2017). Chat-
bots as conversational recommender systems in urban
contexts. arXiv preprint arXiv:1706.10076.
Li, H., Su, Y., Cai, D., Wang, Y., and Liu, L. (2022). A
survey on retrieval-augmented text generation. arXiv
preprint arXiv:2202.01110.
Pantano, E. and Pizzi, C. (2023). Ai-based chatbots in con-
versational commerce and their effects on product and
price perceptions. Electronic Markets.
Perlin, R., Ebling, D., Maran, V., Descovi, G., and
Machado, A. (2023). An approach to follow microser-
vices principles in frontend. In 2023 IEEE 17th In-
ternational Conference on Application of Information
and Communication Technologies (AICT), pages 1–6.
IEEE.
Radford, A. et al. (2019). Language models are unsuper-
vised multitask learners. Technical report, OpenAI
GPT-2.
Singh, A. (2024). The rise of large language models(llms).
Accessed: 2025-01-19.
Times, T. N. Y. (2021). How chatbots are changing cus-
tomer service. Accessed: 2023-10-01.
Weizenbaum, J. (1966). Eliza—a computer program for
the study of natural language communication between
man and machine. Communications of the ACM,
9(1):36–45.
Wolf, T. et al. (2019). Transformers: State-of-the-art natural
language processing. In Proceedings of the 2020 Con-
ference on Empirical Methods in Natural Language
Processing: System Demonstrations (EMNLP 2020).
Zhao, P., Zhang, H., Yu, Q., Wang, Z., Geng, Y., Fu, F.,
Yang, L., Zhang, W., Jiang, J., and Cui, B. (2024).
Retrieval-augmented generation for ai-generated con-
tent: A survey. arXiv preprint arXiv:2402.19473.
Zhao, W. X., Zhou, K., Li, J. Y., Tang, T. Y., Wang, X. L.,
Hou, Y. P., Min, Y. Q., Zhang, B. C., Zhang, J. J.,
Dong, Z. C., Du, Y. F., Yang, C., Chen, Y. S., Chen,
Z. P., Jiang, J. H., Ren, R. Y., Li, Y. F., Tang, X. Y.,
Liu, Z. K., Liu, P. Y., Nie, J. Y., and Wen, J. R. (2023).
A survey of large language models. arXiv preprint
arXiv:2303.18223.
ICEIS 2025 - 27th International Conference on Enterprise Information Systems
1084