About
Experienced Compliance Capital Markets Associate at J&E Davy, specializing in regulatory controls, trade surveillance, and process automation. Second Class Honours graduate in Business Studies with a Data Analytics specialism from DCU, with advanced skills in SQL, Python, and Power BI. Proven track record in automating compliance workflows, managing staff dealing, and ensuring market integrity through rigorous monitoring and reporting.
Recognized for strong leadership, cross-functional collaboration, and managing complex operations for international clients. Multilingual (English, Romanian, Russian), highly adaptable, and committed to continuous professional development—currently aspiring to pursue the CFA.
Ready to leverage analytical insight, client focus, and business growth within a dynamic financial environment. DownloadMy Skills
Web Scraping
PowerBi
Python
Machine Learning
SQL
Javascript
What i do
Web Scraping
With expertise in web scraping, I have successfully utilised tools such as Beautiful Soup and other techniques in my recent projects. Notably, I have contributed to the development of the Daft Data Analysis App and the Telegram Bi Bot, showcasing my proficiency.
Machine Learning
I have gained extensive experience in implementing various machine learning algorithms, including logistic regression, decision trees, random forests, gradient boosting, naive bayes, and multi-layer perceptrons. I have also leveraged my skills in creating a BERT Transformer Model for email classification.
Data Analysis
I have proficiently utilised a range of data analysis tools, including Power BI and Tableau, to extract valuable insights from diverse datasets. With expertise in handling large volumes of data, I have effectively employed an extensive array of chart types to visually communicate key findings.
0
Lines of Code Written
0
Visualizations created
0
Seconds on this site!
What are you waiting for?
Blogs
The Psychology of Data Visualization: How to Present Data that Persuades
The principles and techniques that underpin the creation of persuasive and effective visuals.
Innovative Strategies for Big Data Management and Analytics
The article explores innovative strategies for managing and analyzing big data efficiently.
Confusion Matrices and Classification Reports
Confusion matrices and classification reports help you understand how well your machine learning model classifies things.
Media
Podcasts
Books
Hands-On Machine Learning with Scikit-Learn, Keras & TensorFlow by Aurélien Géron
The Elements of Statistical Learning by Trevor Hastie, Robert Tibshirani, and Jerome Friedman
Pattern Recognition and Machine Learning by Christopher Bishop
More about me
I have been playing tennis for over three years, and it has quickly become my passion. Through dedicated practice and competition, I have developed a strong appreciation for the sport’s strategy, agility, and mental challenges.
Tennis has allowed me to connect with others, build discipline, and foster teamwork. The thrill of a well-played match and the satisfaction of improving my skills are what drive my enthusiasm for the game.
Much like analyzing data to uncover insights, I enjoy the process of breaking down the complexities of tennis—anticipating shots, refining technique, and adapting to each opponent. The court has become my arena for growth, learning, and personal achievement.
Tennis is an integral part of my life, providing a dynamic outlet for my competitive spirit and a way to express my determination and perseverance.
Contact
Email Classification
Phishing Email Classification with Machine Learning
This project tackles the critical challenge of email phishing detection. It utilizes various machine learning models, including the powerful pre-trained BERT (Bidirectional Encoder Representations from Transformers) model, to distinguish between phishing (spam) and legitimate (non-spam) emails.
Combating Phishing Attempts:
The script equips you with a comprehensive toolkit for evaluating and comparing the performance of different machine learning models on the task of email classification. These models include:
- Logistic Regression
- Decision Trees
- Random Forests
- Gradient Boosting
- Naive Bayes
- Multi-Layer Perceptrons
By analyzing the performance of these models, you can identify the most effective approach for accurately classifying emails and protecting users from phishing scams.
Project Features:
Data Preparation:
The script streamlines data preparation by efficiently loading and preprocessing email data from a CSV file. This includes tasks like label encoding and splitting the dataset for training and evaluation.
Custom Email Dataset Class:
For seamless integration with the BERT model, a custom EmailDataset class is implemented. This class handles email data tokenization, ensuring the data is prepared for the BERT model's input requirements.
Model Comparison and Evaluation:
A core aspect of the project is the ability to train and evaluate various machine learning models on the email classification task. The script facilitates this process for Logistic Regression, Decision Trees, Random Forests, Gradient Boosting, Naive Bayes, and Multi-Layer Perceptrons.
BERT Model in Action:
The project leverages the power of BERT, a state-of-the-art pre-trained model for natural language processing tasks. The script incorporates PyTorch and the Hugging Face Transformers library to train and optimize a BERT model specifically for email classification.
Evaluation Metrics:
To assess the performance of each model during training, the script calculates critical evaluation metrics. These metrics include training and validation loss, accuracy, precision, recall, and F1-score.
Making Predictions:
The script provides a practical predict function. This function allows you to classify new emails as phishing or safe using the trained models. This enables real-time email filtering and protection against phishing attempts.
Saving Trained Models:
For efficiency and reusability, the script saves the trained models' states. This eliminates the need for retraining every time you want to classify new emails. Simply load the saved models and use the predict function for quick and reliable email classification.
Understanding Model Predictions:
The project incorporates the LIME (Local Interpretable Model-Agnostic Explanations) library. This library provides explanations for the model's predictions on specific emails. By understanding the factors influencing the model's decisions, you can gain valuable insights into how the model identifies phishing emails.
License:
This project is licensed under the MIT License, making it freely available for use and modification. See the LICENSE file for details.
Authors:
Bohdan Boiprav & Catalin Bondari
Telegram BI Bot
Telegram Business Intelligence Bot
Python Telegram bot is a versatile tool that leverages web scraping, API integration, and database management to fetch and deliver real-time data to users. It operates on a scheduled basis using cron tasks to ensure that the information provided is always up-to-date. The bot is hosted on KOYEB Servers, utilizing a PostgreSQL database for data storage.
Features:
Usage:
To begin using our Python Telegram bot, follow these simple steps:
Authors:
Bohdan Boiprav & Catalin Bondari
Time Series Features
Time Series Features Project: Activity Recognition with Accelerometer Data
This project tackles the exciting challenge of recognizing activities using time-series data collected from accelerometers. It accomplishes this by extracting features from segmented data and employing machine learning models for classification.
Empowering Activity Recognition:
If you're looking to leverage the power of accelerometer data to identify physical activities, this project provides a solid foundation. It equips you with the tools to translate raw sensor data into meaningful insights about human movement.
Getting Started:
To get up and running, ensure you have the following Python libraries installed:
- pandas (https://pandas.pydata.org/)
- numpy (https://numpy.org/)
- matplotlib (https://matplotlib.org/)
- scikit-learn (https://scikit-learn.org/)
With these libraries in place, you're ready to explore the script and unlock the potential of accelerometer data for activity recognition.
Project Core Functionalities:
Data Loading:
The script assumes your data is organized within a folder named "data" inside the project directory. It iterates through subfolders within "data," reading all CSV files. Each CSV file is expected to contain two crucial pieces of information:
- Accelerometer data (X, Y, and Z axes) capturing movements.
- A designated column named "type" that specifies the activity type for each data point.
Data Preprocessing:
The script incorporates two key preprocessing steps to prepare the data for machine learning models:
- Concatenation: All data points from the CSV files are merged into a single Pandas DataFrame for efficient processing.
- Standardization: A technique known as standardization is applied using a StandardScaler object. This ensures the accelerometer data (X, Y, and Z axes) is on a common scale, leading to better model performance.
Feature Engineering:
This stage involves transforming the raw accelerometer data into features that are more suitable for machine learning models to understand activity patterns.
- Segmentation: The script divides the data into fixed-length segments, creating windows of accelerometer data for analysis.
- Feature Extraction: To capture the characteristics of each segment, the script calculates a comprehensive set of statistical features. These features include mean, standard deviation, median, minimum, maximum, range, and interquartile range (IQR) for each accelerometer axis (X, Y, and Z).
- Feature Storage: The extracted features are meticulously stored in a dictionary and subsequently appended to a list named "segments." Additionally, the activity type corresponding to the first element within each segment is accumulated in a separate list named "labels."
- Feature DataFrame: Finally, a DataFrame named "features_df" is constructed from the "segments" list. The "labels" list is incorporated as a new column named "type" within this DataFrame, providing a clear association between features and activity types.
Machine Learning Models:
The script evaluates the performance of two machine learning models that are well-suited for activity classification tasks:
- Random Forest Classifier: This is a popular ensemble method known for its effectiveness and robustness in various classification problems.
- Linear Support Vector Classifier (SVC): This is a powerful linear classifier that excels at separating data points into distinct classes, making it a strong candidate for activity recognition.
Author:
Catalin Bondari
Data Analysis App
Daft.ie Property Data Analysis App
This project is a web application designed to help users analyze and visualize property data from Daft.ie, a popular Irish property website.
Features:
- Scrapes and cleans property listings: Uses Python libraries to automatically collect and process data from Daft.ie.
- Secure user authentication: Login and registration system ensures user privacy.
- Data storage and visualization: Stores scraped data and offers an interactive dashboard for generating charts and graphs based on user-specified filters.
- Search and filter options: Allows users to search for properties by location and price range.
- Real-time parsing: Parses data directly from Daft.ie with a click of a button.
- CSV cleaning tool: Helps users clean downloaded CSV files by removing unwanted values.
Benefits:
- Effortless data collection: Saves time and effort compared to manual data gathering.
- Clean and organized data: Provides a structured dataset for analysis.
- Visual insights: Enables users to easily identify trends and patterns in the property market.
- Informed decisions: Offers valuable data for property investors, renters, and market researchers.
Target Users:
- Real estate investors
- Property renters
- Market researchers
- Anyone interested in Irish property data
Technology Stack:
- Python (programming language)
- BeautifulSoup and Requests (web scraping libraries)
- Tkinter (GUI development library)
Open Source:
This project is licensed under the MIT License, making it freely available for use and modification.
Author:
Catalin Bondari
Power BI Analysis
Business Analytics Overview of Blackbird's Sales Data
This project provides a business analytics dashboard for Blackbird, a company seeking to gain valuable insights from its sales data. It empowers Blackbird to analyze various aspects of its sales performance and make data-driven decisions for business growth.
Unveiling Sales Trends:
The dashboard offers a detailed breakdown of individual orders, including customer information, location data, order amounts, sales figures, product quantities, and profit margins. This comprehensive view allows Blackbird to track sales trends, identify top-selling products, and monitor profitability across different regions and timeframes.
Geographic Reach and Preferences:
Interactive geographic visualizations are key features of the dashboard. A colored sales volume map of the United States pinpoints Blackbird's strongest and weakest performing states. Additionally, a filled map reveals the most popular product sub-categories within each state, providing valuable insights into regional consumer preferences.
Optimizing Business Operations:
Beyond sales figures, the dashboard delves into operational efficiency. It offers various visualizations like:
- Q&A functionality for quick data exploration
- Cards highlighting top-selling products by category
- Charts displaying the most commonly used shipping methods
- Analysis of how average discount rates impact sales
These insights empower Blackbird to optimize logistics, pricing strategies, and promotional campaigns, ultimately enhancing both efficiency and profitability.
Understanding Your Customers:
The dashboard captures customer data (ID, name, location). This data allows Blackbird to segment its customer base and tailor marketing and sales strategies accordingly. By understanding customer behavior and preferences, Blackbird can cultivate stronger customer relationships and foster loyalty.
Data-Driven Decision Making:
Overall, this interactive Power BI dashboard equips Blackbird with a comprehensive view of its sales data. This facilitates data-driven decision making, strategic planning, and continuous improvement across various aspects of Blackbird's business operations.
Revolut Attrition
Revolut Customer Churn Analysis Project
This project tackles customer churn prediction for Revolut, a popular financial technology company. By analyzing their customer churn dataset, the project aims to develop models that can identify users at risk of leaving the platform.
Understanding Customer Behavior:
The analysis starts by exploring the provided customer churn dataset. This data includes demographics, account details, and most importantly, a churn status indicator for each customer.
Data Visualization:
To gain insights into customer behavior, the project utilizes various visualizations to explore the distribution of data points. These visualizations include distributions of factors like:
- Customer age
- Gender
- Number of dependents
- Education level
- Marital status
- Income category
- Card category
- Months as a Revolut customer
- Total number of accounts held
- Months of inactivity in the past year
- Credit limit
- Total transaction amount
- The proportion of churning vs. non-churning customers
Data Preparation:
Before using the data for model building, several preprocessing steps are applied:
- Categorical features, like education level, are converted using one-hot encoding.
- SMOTE (Synthetic Minority Oversampling Technique) is used to address the imbalanced class distribution, where churners might be a smaller group compared to non-churning customers.
- Principal Component Analysis (PCA) helps reduce the dimensionality of the data, especially after one-hot encoding, which can create many new features.
Building Predictive Models:
The project evaluates two machine learning classification models - Random Forest and Support Vector Machine (SVM) - to predict customer churn. These models are assessed using a technique called 5-fold cross-validation, ensuring reliable evaluation.
Model Performance:
The F1 score is used as the primary metric to compare the performance of the two models in predicting customer churn.
License:
This project is licensed under the MIT License, making it freely available for use and modification. See the LICENSE file for details.
Authors:
Bohdan Boiprav & Catalin Bondari