Data Scientist, Technical Writer

Project 1: Intelligent Access Control Using Machine Learning

Technology Used: Azure Active Directory , C#, Python, Keras, TensorFlow,
Deep Learning, Flask, Anomaly Detection

Objective : This project aims to enhance the security and efficiency of access
control in the organization's IT infrastructure by using machine learning
algorithms to make intelligent access decisions based on user behavior and
historical access patterns.

Overview : Traditional access control systems rely on fixed rules and static
policies to determine user access rights, which can be rigid and may not adapt




well to dynamic user behavior. By integrating machine learning into the IAM
system, we created an intelligent access control solution that continuously
learns from user interactions and adapts access decisions accordingly.

By leveraging cutting-edge AI algorithms, IAM (Identity and Access
Management) systems can detect anomalies in user behavior and pinpoint
potential threats before they cause any damage. An example of this would be
the ability of an AI-based IAM system to analyze user login behavior, such as
time, place, and even user actions, and mark any dubious activity.

Key Components and Steps :

1. Data Collection: The first step was to gather relevant data about user
access patterns, such as login times, accessed resources, geographic
locations, and types of actions performed.
2. Data Preprocessing: Before feeding the data to the machine learning
models, we needed to preprocess and clean the data to remove any
noise or inconsistencies.
3. Feature Engineering: Created meaningful features from the collected
data to help machine learning models identify patterns and anomalies
effectively.
4. Model Selection: Choose appropriate machine learning algorithms that
are well-suited for the access control problem. The model used an
anomaly detection algorithm called Isolation Forest.
5. Model Training: Trained the selected machine learning models using
historical access data to learn normal user behavior and access
patterns.
6. Real-Time Monitoring: Deployed the trained models to work in real
time, continuously monitoring user access requests and generating
access decisions based on their behavior.
7. Dynamic Access Decision: When a user requests access to a
resource, the machine learning system evaluates the request based on
the user's behavior compared to their historical patterns and the
behavior of similar users. If the request deviates significantly from the
norm, the system may prompt for additional authentication or deny
access.
8. Feedback Loop: The system continuously learns and improves its
access decision-making capabilities by incorporating feedback from
access outcomes and user feedback. This feedback loop ensures that
the system becomes more accurate and adaptive over time.
9. Adaptive Privileges: The system can dynamically adjust user
privileges based on their behavior over time. If a user consistently
exhibits responsible behavior, they may be granted more privileges,
while risky behavior may lead to temporary restrictions or increased
scrutiny.




Benefits :

* Improved Security: The integration of machine learning allows the IAM
system to identify potential threats and anomalous behavior more
effectively, reducing the risk of unauthorized access and data breaches.
* Enhanced User Experience: By adapting access controls based on
user behavior, the system can provide legitimate users with a seamless
and frictionless user experience while maintaining a strong security
posture.
* Reduced Administrative Overhead: The system's ability to make
intelligent access decisions reduces the need for manual access
management, saving time and effort for administrators.
* Continuous Learning: The IAM system becomes more effective over
time as it continuously learns and adapts to changing user behavior and
emerging access patterns.

I wrote a brief blog on this topic as well.
https://blog.paperspace.com/anomaly-detection-isolation-forest/

Project 1: Object Detection and Image segmentation Machine Learning
for editing visually captivating video Advertisement

Technology Used: Python, Computer Vision, Convolutional Neural Networks
(CNN), Object Detection, Image Segmentation, OpenCV, TensorFlow, Keras,
YOLO, DeepLab, Non-Maximum Suppression, Morphological Operations.

Objective : Creating an engaging and visually captivating video advertisement
for the latest flagship credit card advertising using object detection and image
segmentation techniques.



Key Contributions and Features:

1. Data Collection and Preprocessing: I led the effort to gather a diverse
and extensive dataset of annotated images containing credit cards, the
object of interest.




2. Choose a pre-trained object detection model, YOLOv3, and fine-tune it
using the annotated dataset.
3. Applied the trained object detection model to each frame of the video.
Use the model's output bounding boxes to track the credit card's
position and movement throughout the scenes.



4. Choose a pre-trained image segmentation model, DeepLab, and train it
on images where the credit card needs to be isolated from the
background. This created a pixel-level mask for the credit card.



5. Utilized the object detection output to add dynamic annotations and text
labels that follow the credit card's movement in the video. Applied the
image segmentation masks to create visually appealing effects.

Project Outcomes: Object detection and image segmentation techniques add
a dynamic and modern touch to video editing for video advertisement.

Automated annotations and effects maintain consistency and accuracy across
frames.

Interactive elements create an immersive experience for viewers, increasing
engagement.

Visual enhancements, like background blurring and unique filters, make the
video visually appealing.

Project 2: Natural Language Processing (NLP): Bank Document Analysis
with Large Language Model (LLMs) GPT-3.5-Turbo

Technology Used: Python, NLP, GPT-3.5-Turbo, Chromadb, LangChain,
Gradio, AWS
Objective : Use a large language model like GPT for various bank document
analyses. In this project, I used GPT-3.5-Turbo. It is a large language model
that can used in the banking and insurance sectors. It can help banks and
other financial institutions to extract and classify data from various bank
documents in PDF and other formats with greater accuracy, speed, and
efficiency.
Project high-level steps:

* Split the document data into smaller segments, generated vector
representations of those parts, and built a vector database.




* Defined a prompt template to produce prompts for the OpenAI language
model.
* Called OpenAI API via LangChain
* Deployed the model on AWS and used Gradio for interaction


Project 3: Personalized Credit Card Recommendation System

Technology Used: Python, Collaborative filtering, Content-based filtering,
Django, Dialogue Flow

The Personalized Credit Card Recommendation System is a machine learning
project designed to assist a bank's customers in finding the most suitable
credit card options based on their financial profiles, spending habits, and
preferences. By analyzing customer data and transaction history, the system
aims to provide tailored credit card recommendations that match individual
needs and maximize benefits.

* Collected customer data, including income levels, employment status,
credit scores, transaction history, spending categories, and previous
credit card usage.
* Cleaned and preprocessed customer data to handle missing values,
outliers, and inconsistencies.
* Created features that capture relevant information, such as average
monthly spending in different categories and credit utilization rates.
* Employed clustering techniques to segment customers based on their
financial behaviors and preferences.
* Utilized collaborative filtering to identify credit cards preferred by
customers with similar profiles.
* Applied content-based filtering to match credit card features with
individual customer preferences.
* Integrated the recommendation system into the bank's online banking
portal and mobile app.
* Created a chatbot using Dialog flow and integrated it into the bank's
website. This chatbot recommends the credit card using the same
model.




Project 4: Conversational AI - Natural Language Processing (NLP) -
Deep Learning




* In this machine learning project, we aimed to develop an intelligent and
efficient Customer Service Chatbot. Chatbots are often used by
businesses and organizations to automate customer service, sales, and
marketing interactions, as well as to provide 24/7 support to their
customers. They can also be used for personal purposes, such as
entertainment, education, and productivity.

* In this project, we created a Chatbot using Python, Machine learning,
Natural Language Processing or NLP, Keras, and NLTK or Natural
Language Toolkit.

* The model has two Dense layers, where the first layer has 64 neurons
with ReLU activation, and the second layer has ten neurons with
SoftMax activation. The input shape parameter specifies the shape of
the input data. The model is compiled with a categorical cross-entropy
loss function, the Adam optimizer, and the accuracy metric.

* AWS Setup: We utilized AWS services to deploy and manage the
chatbot. This involved setting up an Amazon EC2 instance to host the
model and an Amazon API Gateway for handling incoming chatbot
requests.
* Please read my blog related to this project
https://botbark.com/2023/03/13/chat-bot-using-python-machine-learning-nlp-keras-and-nltk/


Project 5: Text to Video Generation using Deep Learning, PyTorch, and
Python

* Text to Video Generation using Deep Machine Learning with Python,
Diffusers, Transformers, and PyTorch is an advanced project that aims
to automatically create video content from textual input using cuttingedge
deep learning techniques. The project leverages Diffusers and
Transformers based on self-attention mechanisms to generate highquality
and coherent video sequences that correspond to the given
textual descriptions.
* Key Components and Process:
* Text Input: The project starts by collecting textual descriptions or scripts
that will serve as input to the system. These descriptions can be in the
form of natural language sentences or structured instructions.
* Natural Language Processing (NLP): The text is preprocessed using
NLP techniques to clean, tokenize, and extract relevant information.
The goal is to convert the textual input into a format suitable for deep
learning models.




* Feature Extraction: Diffusers are used to convert the preprocessed
text into a continuous representation. Diffusers are probabilistic
generative models that capture the probability distribution over a
sequence of data. They are used to learn a meaningful representation
of the textual input, allowing for a smooth and interpretable flow of
information.
* Transformer-based Video Generation Model: The project utilizes a
deep learning model based on Transformers, specifically designed for
video generation. The Transformer model takes the continuous
representation from the Diffuser and generates video frames from it.
The self-attention mechanism in Transformers allows the model to
capture long-range dependencies between words and generate
coherent video sequences that align with the textual input.




Project 6: Customer Segmentation using Machine Learning -
Unsupervised Learning

* Customer segmentation means grouping your customers
according to various characteristics (for example, grouping
customers by age).

* It's a way for organizations to understand their customers.
Knowing the differences between customer groups makes it
easier to make strategic decisions regarding product growth and
marketing.

* K-Means clustering is an efficient machine learning algorithm to
solve data clustering problems. It's an unsupervised algorithm
that's quite suitable for solving customer segmentation problems.
Before we move on, let's quickly explore two key concepts

* Finding the optimal number of clusters is one of the key tasks
when implementing a k-means clustering algorithm. It's worth
noting that a k-means clustering model might converge for any
value of K, but at the same time, not all values of K will produce
the best model.

* Please read my blog related to this project

https://neptune.ai/blog/customer-segmentation-using-machine-learning


Project 7: Anti-Money Laundering - Anomaly Detection




* From Anti-Money Laundering and bank fraud to preventative machine
maintenance, anomaly detection is an incredibly useful and common
application of machine learning.
* An outlier is nothing but a data point that differs significantly from other
data points in the given dataset.
* Anomaly detection is the process of finding the outliers in the data, i.e.,
points that are significantly different from the majority of the other data
points.
* For this project, we used Isolation Forest. It is a machine-learning
algorithm for anomaly detection.
* It's an unsupervised learning algorithm that identifies anomalies by
isolating outliers in the data. Isolation Forest uses an ensemble of
Isolation Trees for the given data points to isolate anomalies.
* Please read my blog related to this project

https://blog.paperspace.com/anomaly-detection-isolation-forest/

Project 8: Web3 - Time Series - Predict Ethereum Blockchain Price using
Machine Learning

* A blockchain is a computer network that keeps a shared and up-to-date
transactional database. A block is a new set of transactions introduced
into the system.
* The popular blockchain platform Ethereum is open-source,
decentralized, programmable, safe, and scalable.
* The Ethereum blockchain produces a lot of data because of its inherent
transparency and decentralized structure. It is also known as "on-chain
data," and anyone can easily access it.
* This project aims to introduce Ethereum price prediction using on-chain
data in a straightforward manner. In time series analysis, finance, web3,
or blockchain in general, predicting the price of Ethereum can be a
rewarding subject to work on.
* Please read my blog related to this project

https://dhirajkumarblog.medium.com/web3-using-on-chain-analysis-to-predict-priceof-ethereum-blockchain-10c493a4d4e4

* Utilized Python to extract, clean, and manipulate data from various
sources such as databases, APIs, and CSV files for analysis purposes.
* Applied data wrangling techniques to pre-process and transform raw
data into a structured format suitable for analysis, using libraries such
as Pandas and NumPy.
* Developed and implemented data analysis workflows and pipelines
using Python libraries such as scikit-learn, stats models, and
TensorFlow for tasks like regression, classification, clustering, and
natural language processing.
* Conducted exploratory data analysis (EDA) to gain insights, identify
patterns, and visualize trends in datasets using libraries such as
Matplotlib, Seaborn, and Plotly.
* Created and maintained interactive dashboards and visualizations using
libraries like Plotly Dash and Tableau to present data-driven insights
and findings to stakeholders.
* Developed and implemented statistical models and hypothesis tests to
analyze and interpret data, providing actionable insights and
recommendations for business decision-making.
* Collaborated with cross-functional teams to define project objectives,
gather requirements, and ensure that data analysis solutions aligned
with business goals and requirements.
* Worked with stakeholders to identify key performance indicators (KPIs)
and develop metrics for tracking and measuring the success of data
analysis projects.
* Conducted data quality assessments and implemented data validation
techniques to ensure accuracy, consistency, and completeness of
datasets.
* Documented data analysis processes, methodologies, and findings to
support reproducibility, knowledge sharing, and collaboration within the
team and across the organization.