DATA SCIENCE AND ARTIFICAL INTELLIGENCE

INTERNSHIP REPORT

Submitted By

Siva S
A report submitted in partial fullfillment of the requirements for the Award of Degree
of

BACHELOR OF ENGINEERING
IN
COMPUTER SCIENCE AND ENGINEERING

Under supervision of
[Link] LABS - RASA INSTITUTE OF ANALYTICS,CHENNAI

SCHOOL OF COMPUTER SCIENCE AND ENGINEERING - CSE

SRIRAM ENGINEERING COLLEGE, PERUMALPATTU

ANNA UNIVERSITY: CHENNAI 600 025
JULY/AUGUST - 2025
 BONAFIDE CERTIFICATE

DATA SCIENCE AND ARTIFICAL

Certified that this internship report titled
INTELLIGENCE is the Bonafide work of Mr. Siva SB 4th year/ CSE who carried
out the internship under my supervision. Certified further, that to the best of my
knowledge the work reported here in does not form part of any other internship
report or dissertation on the basis of which a degree or award was conferred on
an earlier occasion on this or any other candidate.

Dr. P. MAGESH MS. S. PRAVEENA

School of CSE and Head, Assistant Professor/CSE,
Sriram Engineering College, Internship Coordinator,
Perumalpattu, Thiruvallur Dist – 602 024. Sriram Engineering College,
Perumalpattu, Thiruvallur Dist – 602 024.
 ACKNOWLEDGEMENT

I take this opportunity to express our sincere gratitude to Rasa AI Labs, Chennai, for
granting us the opportunity to undergo our internship and work on the project
CRM-BASED-ECOM, “genius_shoppy.ai”.he guidance, exposure, and
encouragement we received during this period have been invaluable to our
academic and professional growth.

I are profoundly grateful to our Principal, Dr. B.R. Ramesh Bapu, Ph.D., for the
facilities and constant encouragement extended to us throughout the internship.

I extend our heartfelt thanks to our Head of the Department, Dr. P. Magesh, for his
valuable support and guidance in making this internship possible.

I owe special thanks to Mrs. A. Lavanya for her constructive suggestions and
continuous encouragement, which guided us in refining our work.

I also acknowledge with gratitude the support and guidance of our College Internship
Coordinator, Ms. S. Praveena, Assistant Professor, Department of CSE, for
facilitating and coordinating the internship process effectively.

I remain indebted to all our faculty members of the Department of CSE for their
encouragement, and to our friends for their cooperation and support throughout this
journey.

Finally, I thank each other as teammates for the collaboration and shared effort that
made this project a success.

Siva S
112622104020
 Genius [Link]

Abstract:

Genius_Shoppy.AI is an end-to-end CRM-based e-commerce analytics ecosystem that

combines the power of AI, real-time web technologies, and behavioural intelligence to
revolutionize the online shopping experience. At its core, the system is designed to analyse
customer behaviour through intelligent browser extensions that track user interactions across
major e-commerce platforms like Amazon, Flipkart, Meesho, Zepto, and others. These
extensions communicate with a FastAPI backend that performs advanced customer segmentation
using RFM (Recency, Frequency, Monetary) analysis, enabling targeted marketing and retention
strategies.

The system also integrates powerful explainable AI models using SHAP and LIME to
provide transparency in recommendation logic and user profiling. An analyst dashboard—built
with user-friendly design and professional-grade visualizations—enables business stakeholders
to review insights, tweak strategies, and monitor performance. Web scraping, powered by APIs
like [Link], ensures that real-time product data and best deals are always accessible to users
and analysts alike.

The project is structured to support continuous learning through unsupervised clustering

models that adapt to evolving user behaviours over time. With scalable architecture, modular
codebase, and cross-platform support (browser extension, WebGL, Android, and
desktop),Genius_Shoppy.AI establishes a powerful blueprint for intelligent retail CRM solutions
that combine personalization, automation, and analytics into one seamless system.

Keywords: CRM-Analytics – E-Commerce-Intelligence – RFM-Segmentation – SHAP-

Explainer – LIME-Model – Browser-Extension – Real-Time-Tracking – Product-
Recommendation – Web-Scraping – [Link]-API – Customer-Behavior – Gemini-AI
– Analyst-Dashboard – Explainable-AI – Retail-Personalization – FastAPI-Backend –
Chart-Visualization – User-Profiling – Cross-Platform-Support – Intelligent-Shopping-
System
 TABLE OF CONTENTS

S. No. CONTENTS PAGE NO.

1.

Introduction

2.

Objectives

3.

Literature Survey

4.

System Architecture

5.

Modules

6.

Results & Output

7.

Tools & Technologies Used

8.

Screenshots
1. Introduction

1.1 Background

In the rapidly evolving landscape of online retail, e-commerce platforms generate vast
amounts of user behavior and transactional data. Traditional CRM (Customer Relationship
Management) systems, while effective in managing customer information, often fall short in
delivering real-time insights and actionable intelligence. With the increasing demand for
personalization, deal discovery, and explainable AI recommendations, businesses require
intelligent systems that not only analyze past behavior but also predict future trends and optimize
user experience.

Browser extensions have emerged as effective tools to engage customers in real-time,

while backend analytics engines can segment, analyze, and derive insights from customer data.
Simultaneously, advancements in explainable AI (XAI) frameworks like SHAP and LIME now
offer transparency in model predictions—critical in retail applications where trust and
accountability influence customer decisions. This convergence of CRM, AI, and real-time web
technologies sets the stage for platforms like Genius_Shoppy.AI.

1.2 Purpose

The primary purpose of Genius_Shoppy.AI is to enhance the e-commerce experience for both
users and businesses by:

● Integrating CRM-based analytics directly into web environments through browser

extensions and analyst dashboards.
● Providing data-driven product recommendations and segmentation using the RFM model.
● Making AI decisions transparent using SHAP and LIME for increased stakeholder trust.
● Tracking user behavior to deliver real-time promotional insights and optimal product
deals.
● Offering analysts a GUI-based interface to visualize insights and make informed strategic
decisions.

This system bridges the gap between raw user behavior and intelligent decision-making in retail.
1.3 Scope

The project encompasses a wide technological scope across multiple components:

● Browser Extension: Monitors user search activity on platforms like Amazon, Flipkart,
Meesho, and others, and triggers deal recommendations based on intelligent query
analysis (via Gemini AI).
● Backend (FastAPI): Hosts the logic for RFM segmentation, SHAP/LIME explainers,
and business rule mining, with support for periodic scraping using the [Link] API.
● Frontend Analyst GUI: Displays insights from backend output files (e.g.,
rfm_output.csv, shap_output.json) and renders charts using [Link] with
support for multiple chart types.
● Data Collection & Scraping: Incorporates real-time scraping from multiple e-commerce
platforms to update product listings and identify trending deals.
● Explainable AI: Implements SHAP and LIME for justifying predictions and segment
behaviors.
● Cross-platform Compatibility: Designed to run across web, Android, and desktop
environments using modular components.

By uniting these capabilities, Genius_Shoppy.AI provides a scalable, intelligent, and

explainable e-commerce intelligence framework that can be adopted across various retail and
CRM applications.

2. Objectives

The primary objective of the Genius_Shoppy.AI project is to develop a comprehensive,

intelligent e-commerce analytics system that integrates customer relationship management
(CRM) techniques with real-time data retrieval, behavioral analysis, and explainable AI. The
following sub-objectives outline the core components of this system and the strategic goals
behind each feature:
 2.1. Integrate CRM Analytics (RFM) into a Web Extension and
Analyst GUI

A major goal of this project is to implement Recency, Frequency, and Monetary

(RFM) segmentation within both a user-facing browser extension and an internal analyst GUI.
RFM analysis is one of the most widely used CRM techniques to identify loyal, high-value, and
at-risk customers based on their purchase history. By calculating these metrics from transaction
data, the system can segment users into meaningful categories such as loyal customers, frequent
buyers, or dormant users.

The integration of this model into a browser extension ensures real-time feedback and
product suggestions for users based on their behavior. Simultaneously, an analyst dashboard
allows business users to explore these RFM outputs visually, generate reports, and strategize
accordingly. The goal is to tightly couple backend analytics with frontend visualization to
support both customer engagement and business decision-making.
By automating the process of price comparison, discount tracking, and availability
checks, Genius_Shoppy.AI empowers both customers and analysts with current market insights.
The scraped product data is dynamically linked with customer behavior insights to recommend
the best available deals based on relevance, user preference, and purchase potential.

2.2. Provide Explainable AI Using SHAP and LIME for Customer

Behavior

Another key objective is to enhance the transparency and interpretability of AI-driven

insights using explainable AI (XAI) tools. The system leverages SHAP (SHapley Additive
exPlanations) and LIME (Local Interpretable Model-agnostic Explanations) to explain how
machine learning models make predictions regarding customer segments, behavior patterns, or
purchase likelihood.

SHAP and LIME generate human-readable explanations, allowing analysts to understand

which features (such as frequency of purchases, time since last purchase, or cart abandonment)
are influencing decisions. This builds trust in the system, improves debugging and auditing,
and ensures that the system's predictions align with business logic and ethical considerations.

2.3. Enable Business Rule Mining from Customer Segments

Beyond statistical models, the system also supports rule-based reasoning. Business rule
mining is implemented to extract if-then logic from customer behavior and RFM clusters. For
example: “If a customer has not purchased in 60 days but has a high monetary value, send re-
engagement offers.”

These rules help marketers and CRM professionals automate campaigns, prioritize
outreach, and design customized interventions for different customer segments. The extracted
rules are displayed within the analyst GUI for human validation and can be refined for real-time
deployment.

Together, these objectives aim to build a holistic platform that blends CRM logic, real-time
e-commerce intelligence, AI explainability, and business strategy into a seamless system, thereby
improving customer experience and optimizing business outcomes.

3. Literature Survey

The integration of CRM analytics, real-time product tracking, and explainable AI within
browser extensions and analyst dashboards is a growing field within e-commerce technology.
Several tools and platforms currently attempt to offer partial or complete solutions. This section
reviews prominent systems and compares them with the proposed Genius_Shoppy.AI in terms of
features, architecture, and functionality.

3.1 Browser Extensions for E-Commerce Intelligence

Honey (Acquired by PayPal)

Honey is a well-known browser extension that tracks product prices across multiple e-
commerce sites and applies discount coupons at checkout. It helps users save money through
automated deal discovery and coupon application. However, Honey primarily focuses on price
savings and does not offer any personalized recommendations, CRM integration, or user
segmentation.
 Buyhatke

Buyhatke is an Indian browser extension that provides price comparison, price graphs,
and deal alerts. It also scrapes data from e-commerce websites to notify users about the best
time to buy. While it offers deeper analytics than Honey, it lacks AI-based behavioral
modeling, RFM analysis, and explainable insights. Its functionality is user-centric with no
backend integration for businesses or analysts.
Keepa

Keepa is a browser extension that focuses on price tracking for Amazon products. It
offers detailed price history charts and stock availability information. However, it does not
include customer profiling, behavioral segmentation, or CRM elements. It also does not support
integration with analyst dashboards or business logic automation.

InvisibleHand

This extension provides real-time price comparison and alerts when a product is
available cheaper elsewhere. While efficient in improving purchase decisions, it does not track
user behavior or perform any form of data analytics beyond pricing.

3.2 Analytics Dashboards in CRM and E-Commerce

Google Analytics + Enhanced E-Commerce

Many e-commerce platforms use Google Analytics with Enhanced E-Commerce plugins
to track customer behavior, transactions, and conversion rates. Although powerful, this tool
requires manual integration and does not support explainable AI or business rule generation.
Additionally, it lacks direct scraping capabilities and browser-based engagement features.

Enterprise-grade platforms like Zoho and Salesforce provide CRM tools with
segmentation, reporting, and sales forecasting. However, these systems are often subscription-
based, complex, and not designed for real-time consumer-side interactions via browser
extensions. They also typically require substantial customization to support XAI models like
SHAP and LIME.
3.3 Comparison with Genius_Shoppy.AI

Honey Buyhatke Keepa Google Zoho Genius_Shoppy.

Feature
Analytics CRM AI

Price Tracking Yes Yes Yes Limited No Yes

Coupon Finder Yes Yes No No No No

RFM Segmentation No No No Partial Yes Yes

Explainable AI
No No No No No Yes
(SHAP/LIME)

Business Rule Mining No No No No Yes Yes

Browser-Based Behavior
Yes Yes Yes No No Yes
Tracking

Analyst Dashboard
No No No Partial Yes Yes
Integration

Web Scraping
No Yes Yes No No Yes
Integration

Free and Open System Yes Yes Yes No No Yes

3.4 Summary

While existing browser extensions and analytics tools address specific components such as
price tracking or customer segmentation, they generally lack integration across behavioral
tracking, backend analytics, real-time scraping, and explainable AI. Genius_Shoppy.AI bridges
this gap by combining CRM-driven segmentation (RFM), explainable AI (SHAP,
LIME),web scraping ([Link] API), and real-time user interaction via browser
extensions. This unified approach provides a holistic and scalable solution that benefits both
customers and business analysts.

4. System Architecture

The system architecture of Genius_Shoppy.AI is modular, scalable, and designed for

seamless integration between user-side behavior tracking, backend analytics processing, and
business intelligence visualization. It follows a three-tier structure—Backend, Data Pipeline,
and Frontend—ensuring separation of concerns, ease of deployment, and maintainability across
platforms such as Web, Android, and desktop.

4.1 Requirements
● fastapi
● uvicorn
● pandas
● scikit-learn
● matplotlib
● seaborn
● shap
● lime
● joblib
● python-multipart
● python-dotenv
● aiofiles
● jinja2
● numpy
● pydantic
● psutil
 4.2 Backend Architecture

The backend is developed using FastAPI, a high-performance Python framework that

supports asynchronous execution and RESTful endpoints. It acts as the core processing unit of
the system, exposing a suite of API endpoints that serve both the browser extension and the
analyst GUI.

● RFM Endpoint: Processes transactional data to generate RFM metrics (Recency,

Frequency, Monetary). Outputs are stored in rfm_output.csv for further analysis.
Coding:
def get_rfm_clusters(df):
df['TotalPrice'] = df['Quantity'] * df['Price']
rfm = [Link]('Customer ID').agg({
'InvoiceDate': lambda x: (pd.to_datetime('now') - pd.to_datetime([Link]())).days,
'Invoice': 'nunique',
'TotalPrice': 'sum'
}).reset_index()
[Link] = ['CustomerID', 'Recency', 'Frequency', 'Monetary']

kmeans = KMeans(n_clusters=4, random_state=42)

rfm['Cluster'] = kmeans.fit_predict(rfm[['Recency', 'Frequency', 'Monetary']])
return rfm
● SHAP and LIME Endpoints: These endpoints explain model predictions and customer
behaviors using SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable
Model-agnostic Explanations). The outputs are JSON-formatted and used to display
feature influence in the analyst GUI.
CODING:
def get_shap_results(target_column: str = "Country", path: str =
DEFAULT_DATASET_PATH):
try:
df = fetch_dataset(path)
result = explain_shap(df, target_column)
return {
"status": "success",
 "type": "shap",
"target": target_column,
"feature_names": [Link]("feature_names", []),
"mean_abs_shap_values": [Link]("mean_abs_shap_values", [])
}
except Exception as e:
return {
"status": "error",
"type": "shap",
"message": str(e)
}

def get_lime_results(target_column: str = "Country", path: str = DEFAULT_DATASET_PATH):

try:
df = fetch_dataset(path)
result = explain_lime(df, target_column)
return {
"status": "success",
"type": "lime",
"target": target_column,
"lime_explanation": result
}
except Exception as e:
return {
"status": "error",
"type": "lime",
"message": str(e)
}
● Business Rule Mining Endpoint: Derives logical rules from customer segments, e.g., “If
Frequency > 5 and Recency < 30, then high-engagement customer.” These rules support
automated decision-making in CRM strategies.
CODING:
def get_rule_insights(path: str = "rfm_output.csv"):
try:
 rfm_df = pd.read_csv(path)
rules = get_business_rules(rfm_df)
return {
"status": "success",
"type": "rules",
"rules": rules
}
except Exception as e:
return {
"status": "error",
"type": "rules",
"message": str(e)
}
Precomputed Data Usage: For performance and deployment efficiency, all outputs are
precomputed and stored (e.g., shap_output.json, [Link]), reducing computation
time during frontend rendering.
CODING:
def load_rfm_output():
path = [Link](BASE_DIR, "rfm_output.csv")
return pd.read_csv(path).to_dict(orient="records")

def load_lime_output():
path = [Link](BASE_DIR, "[Link]")
return pd.read_csv(path).to_dict(orient="records")

def load_business_rules():
path = [Link](BASE_DIR, "business_rules.csv")
return pd.read_csv(path).to_dict(orient="records")
4.3 Data Pipeline

The system incorporates a dynamic and modular data pipeline that automates data
collection, analysis, and model feeding.

● User Behavioral Data: Collected via browser extension, including search terms, product
clicks, and page visit patterns. This data is anonymized and sent to the backend for
analysis.
● Product Data Scraping: Real-time product listings and deal information are fetched
using the [Link] API, eliminating the need for maintaining in-house scrapers. This
scraping runs periodically or on-demand to ensure current data is available for
recommendation logic.
● Data Storage and Access: Preprocessed outputs (RFM segments, SHAP explanations,
business rules) are saved as structured files (CSV/JSON) for both API access and GUI
rendering.
●

4.4 Frontend Architecture

The system features two primary frontend components: a browser extension for customer
engagement and an analyst dashboard GUI for business insights.

Browser Extension

● Built using JavaScript and Chrome Extension APIs.

● Detects user behavior in real-time (searches, product views).
● Sends relevant events to the backend and receives product recommendations.
● Uses Gemini AI API to analyze search intent (e.g., "best laptop under ₹50,000") and
fetches matching deals from e-commerce platforms.
● Displays recommendations and promotional deals within the shopping site interface.

Analyst GUI (Dashboard)

● Developed using HTML, CSS, and JavaScript.

● Supports rendering of RFM segments, SHAP insights, LIME feature explanations, and
mined business rules.
● Integrated with [Link] for dynamic chart generation, supporting multiple chart
 types like bar, pie, line, scatter, and histogram.
● Includes dropdowns for chart type selection and toggling between RFM, SHAP, LIME,
and Rules modules.

4.5 Deployment and Cross-Platform Support

● The entire backend is deployed using [Link], ensuring public API access.
● The browser extension works across Chromium-based browsers on both desktop and
Android (via Lemur or Kiwi browsers).
● The frontend GUI is portable and can be wrapped into an Android APK or WebGL
desktop app using frameworks like Niotron or Electron.

4.6 Summary

The architecture of Genius_Shoppy.AI is designed to be modular, real-time, and easily

extensible. By decoupling components and leveraging APIs like [Link] and Gemini, the
system achieves seamless integration of user tracking, CRM analytics, explainable AI, and
business intelligence—all while maintaining performance and scalability.

5. Modules

The Genius_Shoppy.AI system is organized into distinct functional modules, each

responsible for specific tasks within the analytics and interaction pipeline. These modules work
together to form a cohesive ecosystem that serves both end users (via browser extension) and
business stakeholders (via analyst dashboard).
5.1 RFM Segmentation Engine

The Recency-Frequency-Monetary (RFM) Engine is the core module responsible for

segmenting customers based on transactional data.

● Input: Historical purchase or interaction logs (CSV or JSON format).

● Process: Calculates:
o Recency – How recently a user made a purchase.
o Frequency – How often they make purchases.
o Monetary – Total spending value.
● Output: A preprocessed CSV file (rfm_output.csv) containing user IDs and their
corresponding RFM scores and cluster assignments (e.g., loyal, frequent, at-risk).
● Purpose: Enables business teams to target customer segments with tailored marketing
strategies.

5.2 SHAP & LIME Explainer Module

This module introduces explainability into customer segmentation and AI predictions by

applying SHAP and LIME.

● SHAP (SHapley Additive exPlanations):

o Assigns each feature (e.g., frequency, cart size, recency) an importance value.
o Outputs a dictionary of contributions per feature per customer.
● LIME (Local Interpretable Model-agnostic Explanations):
o Perturbs the input space to understand decision boundaries.
o Useful for interpreting local behavior around a specific prediction.
● Output: Structured as shap_output.json and lime_output.json files.
● Purpose: Helps analysts and decision-makers understand why certain users were placed
in specific segments or received specific recommendations.
5.3 Business Rules Mining Module

This module extracts if-then logical rules from the clustered RFM data or other behavioral
patterns to support automated decision-making.

● Input: Clustered customer data (from the RFM engine) and engagement history.
● Process:
o Applies rule-mining algorithms (e.g., Apriori or threshold-based logic).
o Identifies actionable rules such as:

▪ “If Frequency > 5 and Recency < 30 → Target with Premium Offers”

▪ “If Monetary < ₹100 and Recency > 90 → Re-engagement Email”

● Output: [Link] or readable markdown-style rules shown in the dashboard.

● Purpose: Enables marketing automation, retention strategies, and personalization.
CODE:
high_value = rfm_df[
(rfm_df['Recency'] <= 90) &
(rfm_df['Frequency'] >= 3) &
(rfm_df['Monetary'] > 500)
]

5.4 Browser Extension Module

This frontend module interacts with users directly during their shopping journey.

● Platform: Chromium-compatible (Chrome, Edge, Lemur for Android).

● Functionality:
o Tracks product searches, views, and interaction behaviors.
o Sends behavioral data to the backend for analysis.
o Uses Gemini AI to parse complex queries (e.g., “best gaming phone under
₹30,000”).
o Fetches real-time deals using the [Link] API and displays top matches.
● Purpose: Provides real-time, personalized shopping suggestions based on CRM insights
and live product data.
5.5 Analyst GUI Dashboard Module

The dashboard is the visual analytics layer for business users and analysts.

● Framework: HTML, JavaScript, [Link].

● Components:
o Dropdown to choose between RFM, SHAP, LIME, or Rules.
o Dynamic chart rendering (bar, pie, histogram, line, scatter).
o File loader to fetch precomputed outputs like
rfm_output.csv, shap_output.json.

● Interactivity:
o Allows selection of chart types.
o Enables toggling between different customer clusters and features.
● Purpose: Facilitates decision-making through visual insight, trend tracking, and model
transparency.

5.6 Inter-Module Communication

All modules are connected via well-defined APIs and file-based output sharing:

● Browser extension communicates with backend via FAST APIs.

● Backend shares precomputed JSON/CSV outputs with the GUI.
● Analyst GUI fetches these outputs and renders them in real time.

6. Results & Output

The implementation of Genius_Shoppy.AI has successfully validated the core

functionalities envisioned in the system design. The results demonstrate the seamless integration
of backend analytics, intelligent customer segmentation, and real-time interaction through
browser extensions and dashboards.

6.1 RFM Segmentation Output

 ● The RFM Engine successfully processed transactional datasets and produced a
structured rfm_output.csv file.
● Each customer was assigned Recency, Frequency, and Monetary scores, along with
corresponding cluster labels such as Loyal, Frequent Buyer, Potential Churn, and
Low-Value.
● The segmentation data has been effectively utilized by both the dashboard and the
business rule mining module to drive decision-making.

6.2 Explainable AI Visualizations

● SHAP and LIME explainers generated detailed model interpretation files

(shap_output.json and lime_output.json) to clarify customer behavior patterns.
● SHAP visualizations clearly outlined feature importance such as time since last purchase,
number of sessions, or total cart value.
● LIME explanations offered local interpretability for individual predictions, providing
justification for each customer segment assignment.
● These visualizations were accurately rendered in the analyst GUI using charts for
intuitive understanding.

6.3 Business Rule Generation

● The system mined human-readable business rules from RFM clusters and other user
behavior features.
● Example rules included:

o "If Recency < 15 and Frequency > 5, then Tag as High-Potential Buyer."
o "If Monetary < ₹100 and Recency > 90, flag for Re-engagement."
● These rules were automatically exported in structured JSON format and were displayed
in the dashboard for analyst review and campaign planning.

6.4 Browser Extension Results

● The browser extension successfully captured user search queries and interaction
behavior across supported e-commerce platforms.
● Intelligent parsing using Gemini AI allowed the extension to interpret complex queries
(e.g., "budget smartwatch for fitness under ₹2000") and fetch relevant product categories.
 ● Based on customer profile data received from the backend, the extension promoted
personalized deals and contextual product suggestions directly within the user’s
browsing session.
● This interaction loop between user behavior, backend analytics, and frontend
recommendations validated the full cycle of intelligent CRM integration.

6.5 Analyst Dashboard Insights

● The analyst GUI dynamically rendered charts based on the type of insight selected—
RFM clusters, SHAP feature contributions, LIME explanations, or business rules.
● Charts were generated on-demand using QuickChart APIs, supporting multiple formats
like bar, pie, line, and scatter plots.
● The GUI provided options to switch chart types, filter clusters, and explore customer
behavior visually, offering business users a real-time insight portal.

Summary of Outputs

Component Output Format Purpose

RFM Engine rfm_output.csv Customer segmentation

SHAP Explainer shap_output.json Global feature importance visualization

LIME Explainer lime_output.json Local model interpretability

Business Rules [Link] Actionable campaign logic

Browser Extension Inline HTML Elements Real-time product recommendations

Analyst GUI Interactive Charts Strategic insight and decision support

The system demonstrates functional completeness in integrating AI-driven insights into

real-time user interfaces, thereby fulfilling the project’s technical and business objectives
effectively.
7. Tools & Technologies Used
The development and deployment of the Genius_Shoppy.AI system relied on a combination of
modern web technologies, data science libraries, and cloud services to ensure scalability,
responsiveness, and intelligent automation.

7.1 Backend Technologies

● Python
o FastAPI: For building RESTful APIs with asynchronous performance.
o pandas: Used for data manipulation, RFM calculation, and preprocessing.
o scikit-learn: For unsupervised clustering, SHAP and LIME integrations.

● Gemini AI API
o Enabled intelligent understanding and transformation of complex user search
queries into relevant keywords and product categories.
● [Link]

o Used for cloud deployment of the backend API with auto-scaling and HTTP
access capabilities.

7.2 Frontend and User Interface

● HTML & CSS
o Designed the Analyst GUI and browser extension interface with a clean layout,
responsive elements, and intuitive controls.

● JavaScript
o Integrated with Chrome extension APIs to capture user behavior and render AI-
generated product suggestions in real time.
o Used for dynamic chart rendering and backend data fetches within both the
extension and dashboard.

● QuickChart API
o Powered the dynamic visualization engine in the analyst GUI for rendering bar,
pie, line, scatter, and histogram charts.

7.3 Browser Extension

● Leveraged the Chrome Extensions API to:
o Monitor user search and navigation behavior.
o Display AI-recommended deals and insights as interactive popups.

o Communicate with the backend asynchronously using fetch() and local storage.
7.4 External Services
● [Link] API (Reserved for future use)
o Intended for real-time product data retrieval from e-commerce platforms via
structured API calls.

● GitHub
o Used for total version control management and as a code storage medium for
deployment purposes.

● Gemini API
o Used for enhancing user queries/String and then returns a response as per user's
Product search.

8. Screenshots:
9. Conclusion
The Genius_Shoppy.AI system demonstrates the successful integration of customer
relationship management (CRM) analytics with intelligent product deal promotion in a unified,
cross-platform environment. By combining data-driven decision-making techniques such as
RFM segmentation, SHAP, and LIME explainability, the system empowers businesses to gain
actionable insights into customer behavior and purchasing patterns.
Through a dual-interface model—comprising a browser extension for real-time user
interaction and an analyst dashboard for in-depth data visualization—the system bridges the
gap between end-user engagement and backend intelligence. The browser extension dynamically
interprets user search intent using Gemini AI and presents the most relevant product deals, while
the analyst GUI allows for flexible chart generation and segmentation monitoring.
The backend, built using FastAPI, provides robust, scalable endpoints to serve
precomputed analytics outputs efficiently. The modular architecture ensures ease of maintenance
and future extensibility, including planned integration of real-time data scraping and more
sophisticated behavioral tracking.
Overall, this project presents a scalable and explainable CRM solution that enhances both
business decision-making and customer experience, laying the groundwork for future
enhancements such as recommendation systems, real-time personalization, and integration across
additional marketplaces.
10. Future Enhancements
To improve the robustness, scalability, and accessibility of the Genius_Shoppy.AI system,
several enhancements are proposed for future development:
● Multilingual Query Understanding:
To better serve a diverse user base across global and regional markets, future versions of
the system will incorporate multilingual natural language processing (NLP) capabilities.
This will allow users to interact with the browser extension and receive relevant product
recommendations in their preferred language, improving inclusivity and personalization.
● Real-Time Model Retraining:
The current system relies on precomputed outputs for RFM, SHAP, and LIME analytics.
A future iteration will introduce real-time retraining mechanisms that adapt to newly
collected user behavior data. This continuous learning loop will enable the backend
models to evolve dynamically, ensuring that recommendations, customer segmentations,
and behavioral explanations remain relevant over time.
● WhatsApp-Based Deal Assistant:
To broaden user engagement beyond the browser, a WhatsApp-based assistant is
planned. This assistant will send personalized deal notifications, alerts on price drops,
and analytics summaries directly to users via WhatsApp. This would increase the
system’s reach and real-time usability, especially in mobile-dominant markets.
● Advanced Web Scraping Capabilities:
While the current system uses [Link] for product data collection, future enhancements
aim to implement dynamic, headless browser-based scrapers (e.g., using Playwright or
Selenium) for broader e-commerce platform coverage. This will enable live scraping
from newer or more complex sources, ensuring up-to-date deal information and greater
adaptability to website structure changes.
● Integration of Recommendation Engines:
Future versions may include collaborative or content-based recommendation engines
trained on historical customer interaction data. These engines will complement the
existing analytical insights by offering direct product suggestions tailored to individual
users or segments.
● Security and Compliance Features:
With the increasing importance of data privacy, upcoming updates will also focus on
integrating security protocols such as OAuth for secure user sessions, data anonymization
techniques, and compliance with GDPR and other regional data protection
[Link] enhancements are geared toward evolving Genius_Shoppy.AI into a
full-fledged, intelligent, and user-friendly platform for real-time, personalized e-
commerce interaction and enterprise-grade CRM analytics.
11. References
● Lundberg, S. M., & Lee, S.-I. (2017). A Unified Approach to Interpreting Model
Predictions (SHAP). Proceedings of the 31st International Conference on Neural
Information Processing Systems (NIPS), 4765–4774.
[Link]
● Ribeiro, M. T., Singh, S., & Guestrin, C. (2016). “Why Should I Trust You?”: Explaining
the Predictions of Any Classifier (LIME). Proceedings of the 22nd ACM SIGKDD
International Conference on Knowledge Discovery and Data Mining, 1135–1144.
[Link]
● FastAPI Documentation:
[Link]
● [Link] API Documentation:
[Link]
● [Link] Documentation:
[Link]
● Chrome Extension APIs:
[Link]