INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 735

Conversational AI Personalized Shopping: An Intelligent Chatbot

With Multi-Layered Recommendation in E-Commerce

V.Bhargav.,

D. Krishna Pradhyumna.,

M.Sai Srujan.,

Ms. K Kowsalya

Dept. of Computer Science and engineering , Hindustan Institute of Technology and Science Chennai,

India

Assistant Professor, Dept. of Computer Science and Engineering, Hindustan Institute of Technology

and Science Chennai, India

DOI :

https://doi.org/10.51583/IJLTEMAS.2026.150100064

Received: 13 January 2026; Accepted: 20 January 2026; Published: 07 February 2026

ABSTRACT

Artificial intelligence and full-stack web technology are combined in the solution to the most distressing

issues in online retail. Comfy is the name of this solution. By employing one of the most innovative hybrid

recommender systems utilizing conversational AI chatbots and combining it with intent analytics, purchase

history, and content-based filtering, it assists customers in making purchasing decisions. Since the Comfy

platform is built on the MERN (MongoDB, Express.js, React, Node.js) technology stack, it has two

components: An AI-based product recommendation system, and an shopping portal for customers, and a

admin analytics console for real-time system analytics. This demonstrates how even small and medium-

sized enterprises (SME) can harness the power of advanced AI for hyper-personalization at a highly

reasonable cost using open-source technology and a thoughtfully architected API. Other noteworthy

attributes include the AI-based progressive payment solution with backend security from Razorpay, unique

approaches to the safety and security of e-commerce AI systems, and an adaptive recommendation system

with multi-layered filtering and responsiveness. With the test system to reality computing, Comfy shows the

possibilities of AI and helps e-commerce systems to incorporate AI, making it an essential tool for

companies wanting to create an e-commerce system with AI.

Keywords: Artificial Intelligence, E-commerce Chatbots, Personalized Recommendations, MERN Stack,

Hybrid Filtering, Secure Integration.

INTRODUCTION

The e-commerce in the present digital era has seen a paradigm of change due to the advent of Artificial

Intelligence as an enabling factor that increases User Engagement and personalization of the shopping

experience. However, as functional they may be, the Classical e-Commerce Portals have gradually shown

signs of deficiency in User Interface and intelligence in finding products, giving way to a huge mismatch

between Customer Demands and the availability of Technology to deliver these. The present day solution for

addressing various query-related tasks has different functional blocks in chatbots, while recommendation

algorithms run in a split manner, leading to the lack of integration of User Engagement based on chat or

online conversations.

As can be shown in more recent works, the detailed review of Valencia-Arias et al. (2024) on AI

recommender systems and Hassan's work on Consumer Behavior in 2025 appears to indicate that they too

have encountered similar issues. Even the most state-of-the-art advances in complex recommender systems

of AI have not been able to provide reviews pointing out the implementation of such systems in advanced,

comprehensive ecommerce platforms. While most, if not all, existing communication assistant tools have

been recognized, albeit with little sophistication, for their predictive capabilities, the analyses of ecommerce

machine learning tools have acknowledged the lack of studies aimed at the construction of scalable

predictive systems.

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 736

What describes Comfy as a e commerce new paradigm with conversational artificial intelligence and multi-

dimensional personalisation for the first time integrated intent, history, and content based filtering

recommendation system in a conversational interface, from a technology perspective, with MERN stack,

Comfy is the first and, along with complexity, the platform integrates multiple frameworks of commerce

custodianship, active management analytics and personalised commerce and enhanced oversight frameworks.

This means a business for the first time had the opportunity to personalise all the eCommerce features.

LITERATURE REVIEW

The use of artificial intelligence technology in the sector of e-commerce is a field that has witnessed

immense growth and has a great deal of academic interest among the research fraternity in terms of how such

intelligent systems can be applied for the benefit of the user and for the optimal delivery of business

operations. The present literature review has covered the essential research pieces that have been shared for

the use of AI, machine learning, and chatbots in the ecommerce model. The present body of research pieces

has come together to demonstrate a comprehension of developments along with indicating the critical gaps

that have been witnessed in the development stage, and the present research on the Comfy platform seeks to

address this. The present analysis has been conducted with the objective of determining five important pieces

of research that relate to the recommendation systems, user behavior, chatbot utilization, algorithm

implementation, and the cumulative implementation of AI technology in the e-commendation platform.

As mentioned in the reference [1], Valencia-Arias et al. explain that with AI technology, the usage of a

recommender system is suggested on e-commerce sites to further improvise the quality of interaction as well

as decision-making between users. This document will critically analyze the trends and prominent algorithms

and developments in technology with objectives focused on optimization regarding the correctness of these

systems of recommendation. According to them, independently it may be some advancement of complex

algorithms for collaborative filtering algorithms based on the algorithms of content filtering or some

advancement regarding systems of recommendation; there remains an enormous gap between these two

graphical interfaces, especially conversational interfaces. They themselves have recognized some prominent

gaps present in literary works presuming that LLM conversational interface and systems of recommendation

are to be recognized as two different entities among an entire setup of the computer system.

Hassan [2] in his empirical study examines the effect of the results of personal recommendations through the

use of chatbots driven by the power of AI to the results of trust, satisfaction, and loyalty in the online market.

The study launches a quantitative research towards the psychological impact of the results of the power of AI

to the users, who have a positive linkage to the results of the intelligence of the developed system developed

by the power of AI technology. The study only scratches the surface of the current potential in the

implementation of technology to the developed system. The study by Hassan does not address the dimension

of the differentiated treatment of the recommendations to the users of the developed system, whether guests

or registered users.

Lopez et al. [3] have presented a critical literature review on the issue of adoption/implementation as well as

the application of chatbots in an e-commerce environment. They have systematically searched the existing

literature on this topic to synthesize the exiting evidence on the advantages brought by the application of

chatbots for improving the level of customer engagement as well as operational efficiency. In fact, their

synthesis has confirmed the initial basic requirement for the application of chatbots in an e-commerce

environment-providing 24/7 customer support services as well as resolving initial queries. However, upon

conducting a critical literature review, it was demonstrated that there has been limited research work on the

application of the implemented chatbots in the following manners-rule-based chatbots or the retrieval of

FAQs, with very limited research being carried out on their applications related to advanced personalized

product recommendations.

The 2024-2025 preprint [4] describes the design, deployment and evaluation of machine-learned algorithms

for personalized product recommendations in e-commerce systems. It examines the specifics of various ML

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 737

approaches, algorithms and models, and compares the prediction accuracy and computational efficiency for

user preference prediction. The study found that, despite the maturity of the algorithms, there is little to no

published literature describing the integration of these models into an operational e-commerce ecosystem,

incorporating real-time chatbots. The author also describes the absence of real-time production environment

analytics and dashboards for admins to the monitor and visualize live data streams and the performance of

the recommendation algorithms.

The IJSREM publication [5] describes a study predicting the use of generative AI within the MERN stack to

create custom e-commerce systems. It describes the development of an AI platform predicting user behavior

and making personalized recommendations. This study is relevant to my research as it considers the same

technology stack (MERN) as the Comfy platform. One limitation of this study is the primary focus on the

recommendation engine. There is little to no consideration of the use of GPT/LLM-based conversational

systems to enable a dialogue. Also, the study does not address the enterprise-grade administrative analytics

or the comprehensive security layers required for a fully-fledged commercial product, leaving the possibility

for a more robust solution.

The literature shows considerable advancement both theoretically and algorithmically for AI driven

components in e-commerce, particularly in recommendation systems, adoption studies for chatbots, and

machine learning. However, all the studies reviewed exhibit the same consistent gaps as mentioned in the

literature [1-5]: the absence of an integrated, documented, and system that incorporates an intent aware,

conversational AI chatbot alongside a multilayered, personalized recommendation engine, all built within a

secure, full-stack, and scalable MERN architecture. In addition, there is a gap in examining the case of real-

time administrative analytics that track business outcomes and the performance of AI features simultaneously.

These gaps; integration, incomplete-implementation, and analytics, is what the development of the Comfy

platform aims to address, moving beyond the gaps of algorithmic sophistication towards a truly integrated

and intelligent e-commerce ecosystem.

RESEARCH GAP AND CONTRIBUTION

Research Gap

The research available is fragmented and so performing a thorough analysis is difficult. Valencian-Arias et al.

focus on AI systems and recommendation systems, while ignoring other categories such as algorithms and

Interactive Conversational UIs. Something similar holds for Hassan’s \cite{hassan2020artificial} empirical

study. While he discusses and explains the constructive role AI personalization plays in the building of trust

with consumers, he does not locate such personalization within a full-stack system. Of all cited works,

\cite{lopez2020chatbots} to most people at least, most apparently, focuses on the most obvious role of

chatbots. Chatbots, who were not originally designed as Intelligent systems for Adaptive Personalized

Product Recommendations, have traditionally performed, and continue to perform, as the most basic

customer service representative. There is, in fact, a reasonable gap where system integration is not, to a large

extent, present. For the most part, there is no system where a natural language processing (NLP) chatbot is

the front end to a more sophisticated and tiered recommendation system.

Besides the concern involving systems integration, most papers remain silent despite one attempting systems

analysis of the complete design and implementation. The body of literature on ML recommendation

algorithms [4] is poor compared to the model evaluation literature, and it does not discuss the algorithms in

the context of a fully functional and operational transactional real-time inventory, secure payment processing,

and user Business Management System.

IJSREM [5] is probably the most distinguished among the numerous publications on the integration of the

MERN stack, yet it disregards the majority of enterprise systems, including but not limited to a full range of

integrated security, user role-based admin access control, and real-time commercial intelligence and AI

processing of the transactional data. The findings of the two publications leave the complete architecture for

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 738

end-to-end AI-commerce systems undocumented, and they ignore the imperative administrative component

of the data monitoring, optimizing, and securing the intelligent production systems metrics. The Comfy

platform addresses this with secure, seamless, and simple personalized commerce and conversational AI.

Contribution

Due to closing the first ever recorded integration gap in the literature, Comfy can now raise an architectural

concern about integrating a smart assistant tool into a recommendation system. We suggested an integrated

system comprising a smart assistant tool and a multi-tiered recommendation system. The proposed system

entirely fulfills an integration gap in the study by Valencia-Arias et al. [1] as it addresses the implementation

gap in Hassan’s study [2]. The proposed system contains a smart assistant tool, whereby the user’s natural

language question undergoes an intent detection process (detectIntent()) that classifies the user’s request. It

additionally triggers the hybrid recommendation service (getRecommendations()), which, in turn, retrieves

the available product details along with the purchase information provided by the user.

Additionally, Comfy documented a smart e-commerce system, responding to review requests for more

documentation. Unlike the theoretical frameworks of previous works [4] or the overly simplistic abstraction

of MERN integrations [5], we built and documented a full-suite framework: a complete frontend in React-

Redux, a secure backend in Node.js/Express (REST), a fully AI-tuned backend database, MongoDB, and

complete middleware for authentication, validation, and custom error handling. This paper can be used as a

guide for implementing other AI-based systems.

To address the shortcomings of personalization in previously conducted studies, we proposed a novel hybrid

filtering approach. Our approach integrates the best of three distinct methodologies: (1) intent-driven

keyword matching in relation to the current query typed by the user, (2) collaborative filtering based on the

purchasing history of the logged-in user on our website, and (3) attribute-based content filtering.

To fill the gaps in security and administration oversight, particularly missing in the IJSREM study carried

out by [5], Comfy includes a comprehensive, enterprise-class security system. JWT authentication with

HTTP only cookies, role-based administration authentication, meticulous input validation using Express-

validator, along with secure Razorpay payment system integration using cryptographic signatures, provides

a safe space for the AI capabilities to run.

At the same time, an administrative dashboard has been developed that provides real-time analyses of

conventional business data (sales, earnings) and AI-specific activity (engagement, click-throughs). This

incorporates the requirement for a level of integration of analysis that has been suggested as a problem in

conventional research [4].

An example of one of the major innovations developed in this research is the progressive personalization

model that successfully addresses the guest-user paradox. Some systems limit AI functions to users once

they log into their accounts. Comfy's chatbot, however, provides basic intent analysis and general

recommendations to all users. After the user logs into the system, the chatbot tailors these recommendations

based on the user’s unique purchase history, which was not elaborated on in the previous empirical research

work [2].

To summarize, the Comfy system's impact empirically demonstrates the value of system benefits being

realized even prior to value being realized from system benefits. In this instance, we present a ready-made

system that integrates chatbot AI, intelligent commerce, and necessary security, scaling, and management

features to streamline the application of the system. All of this currently facilitates the progress of ancillary

research on the application of AI in e-commerce.

System Overview and Proposed System

Comfy is an innovative, intelligent, and comprehensive e-commerce platform incorporating a

conversational AI assistant along with e-commerce recommendation systems for products, tapping into the

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 739

gaps identified in the current literature of AI-commerce. Valentina’s solution does an efficient job in

upgrading the current method of how chatbots and recommenders work as two individual, autonomous

models. The proposed solution streamlines the interaction process Valentina’s solution does an efficient

job… by allowing the user to communicate with the model in a natural language query and generate a

distinct product recommendation. It allows the users to have an enjoyable, simplified shopping experience

using AI in home decor, as well as a comprehensive BI (business intelligence) model for the administrators

to manage the whole system.

The proposed solution is intended to cope with issues related to boundaries, such as closing gaps in

integration in Valencia-Arias et al. [1] and filling the concept of gaps in the process of implementation in

Hassan [2]. This proposed solution includes only one process. In this process, the process of user queries

undergoes stages of intent, review, and identification of product responses. This proposed solution places

the chat interface above question-answer operations, acting as a cognitive assistant and, therefore, not a

mindless one.

Moreover, Comfy is positioned as a publicly available template because it is built on the MERN stack and it

addresses the lack of fully functional backend solutions for the frameworks in the IJSREM publication \[5\]

and the ML-based algorithms in the research \[4\]. The system encapsulates the critical enterprise features

often overlooked in prototypes, such as a comprehensive admin analytics dashboard, robust analytics, a

strong firewall, personalized layers for guests and registered users, and other custom features. Comfy

exemplifies the operationalization of a safe and secure AI-driven e-commerce application system.

SYSTEM ARCHITECTURE AND METHODOLOGY

The framework of Comfy consists of multiple layers including the Data Layer (Database), the Application

Logic Layer (Backend & AI Services), and the Presentation Layer (Frontend). The Presentation Layer

employs the use of React and the Redux Toolkit creating a responsive single-page application that consists of

product browsing, user authentication, cart functionalities, and the chatbot. Each of these components is

managed through a separate slice (auth, cart, and search) that is further divided into smaller slices. The API

interactions and caching are managed via RTK Query which is responsible for optimizing overall system

performance and user interactions.

Using the Node.js and Express.js frameworks, the system's core is built. The Application Logic Layer

employs the MVC architecture, and provides for separate routing, controller, and middleware components.

The middleware components of interest are protect (JWT Authentication), admin (role based access control),

and validator (request verification). One of the key components is the AI subsystem and its unique

methodology which follows a linear/step-wise process. The chatbot Handler controller receives the user

message and sends it to the detectIntent() function which subsequently calls the getRecommendations()

service. This service uses a hybrid methodology to filter the data it retrieves from the database based on user

intent, and if applicable, on the user’s purchasing history from the Order model.

MongoDB’s adaptability in using different types of data makes it a suitable choice for the Data Layer. The

schema uses embedded documents where relevant (e.g., product reviews) and uses pointers for relationships

(e.g., user pointers in orders). The personalization model methodology relies on the data model and combines

three types of filtering. The first is intent filtering, which is done through pattern matching by using regex on

product name or category from the user's query. The other is collaborative filtering, where the system looks

at the order history of the logged-in user and identifies category orders, and the last is basic content filtering

where the system reorders the results based on assigned ratings. This methodology effectively solves the cold

start and single method issues described in previous works [1], [4].

The last architectural pillar is the integration of external services. The system uses the OpenAI (GPT-4o-mini)

API for generating NL responses, however, due to the system’s prompt, the model is instructed to retrieve

only the products from the list and recommend them to the user. The system also uses the Razorpay API to

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 740

complete the payment, and a secure server-side validation (crypto.createHmac) to ensure that the transaction

is legitimate. This architecture is flattened and service-oriented because of the integration of third-party

services, which increases the vertical scalability and separation of concerns of the system.

Figure 1 Architecture Diagram

Implementation Details

Front-end part here is divided into categories based on the React component at the feature level. This

includes overall web pages such as HomePage with the floating button initiating the functionality of the

chatbot system, ProductPage, CartPage, and then further involving the multi-step process for the ordering

functionality pages such as ShippingPage, PaymentPage, and finally PlaceOrderPage. Further, the

functionality of the chat system here is handled by using the Card component. Card here is fixed at the

bottom page level and is conditionally displayed based on the state 'showChat'. In addition to that, it also has

its own internal state 'chatHistory' that broadcasts messages to the backend component by using the

'useChatbotMutation()' function provided by RTK Query. Further, there is an admin part here that includes

individual layouts based on the split routes such as 'dashboard', 'product management', 'order management',

and finally 'user management', with each component being guarded by using the 'AdminRoute' component.

The API is structured to have backend endpoints (such as /api/v1/products, /api/v1/chatbot). In the

chatbotController.js file, the main AI part is processed through a call to the function detectIntent(message).

This function corresponds to a three-way keyword match classification task, namely GREETING,

RECOMMENDATION, or CHAT. In the GREETING and CHAT categories, the message is piped through

the OpenAI API with the system message asking the API to refrain from making unfocused

recommendations. In other cases, the AI triggers the recommendations in the function getRecommendations()

in the file recommendationService.js. In this function, the first step is to build a database query through the

function Product.find(), including an $or condition with the internal keywords (such as pillow or sofa), and

also filtered by the user's categories in their past purchases if req.user is not null.

Security is a point which the system takes seriously. The user passwords are hashed using the bcrypt

algorithm (salt-Iterations: 10). The token is signed using a secret and stored in HttpOnly cookies,

safeguarding it against theft by XSS attacks. The authentication middleware checks the token, which, if valid,

appends the user object to req.user. All the incoming data is validated against the specified logic using

Express-Validator in the route handlers (e.g., validator.checkLogin). In the paymentController.js, the

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 741

payment process does not require the capture of sensitive data. It pays by placing an order using Razorpay,

then verifies the signature sent back using the secret.

The models in the database are defined through mongoose schemas. The Product schema has a nested review

document. The Order schema is referenced to the User schema with an array of orderItems referencing the

Product schema. This enables optimized population queries, like Order.findById().populate("user", "name

email"). The handling of file upload for the product image is done through the Multer middleware in the

uploadRoutes file, where the image file is stored in the ‘uploads’ directory with a server-accessible URL

stored in the Product document.

EXPERIMENTAL RESULTS AND PERFORMANCE ANALYSIS

Comfy's AI main subsystem went one step further by attempting to close the gaps in chatbot-

recommendation systems. We received 50 queries from users to test the system. The user queries included a

variety of use cases including 'hello', 'find blue cushions', 'suggest home decor', questions directed at the

chatbot (e.g. 'what's the weather today?') and questions aimed at driving the recommendation engine. We

achieved an approximate 86% success rate in determining the intent of the user queries. The

recommendation engine operated correctly in making meaningfully aligned product suggestions to the user

queries (provided the user was logged in to the system and the recommendation engine was plugged into

their purchase history). User feedback was quite positive regarding the integrated design, and the ability to

chat and then be directed to a product was a great improvement on the experiences noted in previous research

initiatives (recent 3 literature reviews).

The system was operationally viable based on performance benchmarks. Average response time of the end to

end chatbot was 1.5 seconds and included intent detection, database query, and optional HTTP call to the

GPT API. This time ensured fluidity of conversation. Database performance was robust; the sub 300ms

response time of the getRecommendations() function was achieved even with a hundreds of products. This

was a compound query on the indexed fields name, category, and user-history sub. The 2 seconds deadline

was met by the admin dashboard, which synthesized and charted the analytics (using Recharts) for real time

monitoring of business + AI performance. The multi layered security implemented was effective as the tests

concluded. Unauthenticated data access to Razorpay was blocked, and the server-side signature was

validated. All transaction related security checks in the sandbox environment were passed.

DISCUSSION

The implementation and development of Comfy illustrate the existence of a possible implementation of an

overall stacked AI commerce solution to address the split implementation problem and the integration gaps

in the literature [1] [3] [4] respectively. The implementation of the conversational interface with intent

detection based on fixed external LLM APIs demonstrates how they can utilize a natural and efficient

frontend for a database-driven multi-source recommendation engine. This implementation illustrates the

validity of the hypothesis of the improvement of the user experience through the integration of the user and

the shopping-related conversation for the introduction of the AI personalization component. There are, of

course, trade-offs in the implementation of the proposed solution. The implementation of the natural

language processing function as a keyword-driven intent detection solution serves as an exemplary case of a

natural language comprehension solution. This serves to illustrate limitations in understanding complex

queries in a particular domain. The implemented solution is efficient in taking advantage of the already

developed hybrid recommendation engines.

Also, the development highlights the importance of non-functionals in prototype development. The system's

trust necessitated the incorporation of a working security system. This involved ment authen- tication, input

sanitization, and secure payment mechanisms. The admin dashboard in the prototype, in contrast to the

current literature's reporting function, was instrumental to the system's oversight. It gave the admins the

ability to follow and align the chatbot functions to the payments in order to complete the AI's performance

INTERNATIONAL JOURNAL OF LATEST TECHNOLOGY IN ENGINEERING,

MANAGEMENT & APPLIED SCIENCE (IJLTEMAS)

ISSN 2278-2540 | DOI: 10.51583/IJLTEMAS | Volume XV, Issue I, January 2026

www.ijltemas.in Page 742

assessment. This background work offers a comprehensive outlook and concludes that, in the realm of e-

commerce, the AI technology's top performance is heavily influenced as much by the ingenuity of the

algorithm as by the parameters of security, oversight, and overall manageable system design.

CONCLUSION AND FUTURE WORK

This research work introduces the Comfy solution, an intelligent e-commerce system. It successfully

integrates a conversational chat bot with a hybrid recommendation engine in a secure MERN technology

stack.

Comfy fills a gap as it acts as a working example implementation of an e-commerce solution based on

current theoretical advances in the field of AI. Its architecture blueprint proves the intent of personalization

through the discovery of products, how security features need to play a part in all interactions with artificial

intelligence, and the significance of analytics in the administration of an intelligent e-commerce solution.

For future developments, priorities include further intelligence and scalability improvements to the system.

Currently, it is necessary to upgrade the intent detector from keyword recognition to a fine-tuned transformer

to handle intricate queries. The recommendation system may also benefit from including a collaborative

filtering component to identify trending products by evaluating anonymized purchase data from every

customer. Adding caching (Redis) to process frequent queries and product information will further optimize

scalability improvements. Further development ideas include incorporating support for voice interfaces and

an augmented reality component to visualize products in spatial arrangements and incorporating predictive

analytics into the admin panel to provide insights into inventory and demand forecasting. This is made

possible by designing Comfy with future e-commerce intelligent systems development at its core.

REFERENCES

1. Alejandro Valencia-Arias, Hernán Uribe-Bedoya, Juan David González-Ruiz, Gustavo Sánchez

Santos, Edgard Chapoñan Ramírez, Ezequiel Martínez Rojas, Artificial intelligence and

recommender systems in e-commerce. Trends and research agenda, Intelligent Systems with

Applications, Volume 24, 2024, 200435, ISSN 2667-3053,

https://doi.org/10.1016/j.iswa.2024.200435.

2. Hassan, N., Abdelraouf, M. & El-Shihy, D. The moderating role of personalized recommendations in

the trust–satisfaction–loyalty relationship: an empirical study of AI-driven e-commerce. Futur Bus

J 11, 66 (2025). https://doi.org/10.1186/s43093-025-00476-z.

3. Gamboa Cruzado, Javier & Menendez-Morales, Christopher & López-Goycochea, Jefferson & Alva

Arévalo, Alberto & Vargas, Caleb. (2023). USE OF CHATBOTS IN E-COMMERCE: A

COMPREHENSIVE SYSTEMATIC REVIEW. Journal of Theoretical and Applied Information

Technology. 101. 1172-1183.

4. Patil, Dimple. (2024). Artificial intelligence in retail and e-commerce: Enhancing customer

experience through personalization, predictive analytics, and real-time engagement.

5. Yadav, Abhishek & Vishwakarma, Suraj & Sharma, Siddhant & Yadav, Shivam & Yadav, Saurabh.

(2025). Integration of Generative AI in Mern Stack for Personalized E-Commerce Systems.

INTERNATIONAL JOURNAL OF SCIENTIFIC RESEARCH IN ENGINEERING AND

MANAGEMENT. 09. 1-9. 10.55041/IJSREM52982.