PROJECTS
My data science skillset, which includes analyzing complex data sets, building predictive models, and creating data visualizations, is complemented by my experience in software engineering, where I have developed software applications, designed and implemented algorithms, and worked with different programming languages and tools. This unique combination of skills allows me to develop innovative solutions to complex problems, and makes me a valuable asset to any organization that is looking to leverage data science and software engineering to drive innovation and growth.
Predictive Modelling of Commodities
Skills: Research and analysis, Data extraction and cleansing, Predictive modeling, Time Series Forecasting, Machine Learning, SQL and NoSQL databases.
Data Science Intern at Hatch Quarter
On the dashboard, the user can select the features from the list, by default the dashboard displays the best features that were selected by the Boruta SHAP feature selection algorithm.
The number of estimators, learning rate, max depth, eta, gamma, and subsample may all be changed by the user . Each parameter has a short explanation and a range of values that can be passed below the input field.
When the user clicks the Run Model button, all the features and parameters are sent to the server to fit the model to the given features.
After the model has completed its run, the metrics of the new model and the prediction graph are updated.
Injury Prediction of Atheletes
Data Science Intern at Tri-Alliance
Skills: Data collection, Web services, MongoDB, Data quality, Data Imputation techniques, Exploratory Data Analysis, Machine Learning, Deep Learning and Data Visualization.
The dashboard has four views, and each view will show three plots. The four views are for swimming, cycling, running and overall and are selected using a dropdown menu.
Training IMPulse (TRIMP), uses the weighted increase in heart rate (HR) to quantify load. For instance, the same distances travelled in swimming and cycling are very different workloads placed on the body. TRIMP is used to generalise all the three activities.
Personal wellness factors, including stress levels and sleep quality and duration have moderating effects on recovery. And energy levels and muscle soreness can provide indications of how well the athlete is recovering.
Acute Chronic Workload ratio is a concept developed by sports scientist and is a measure of how quickly the load on the athlete is changing. It is the ratio of the acute load to the chronic load where the acute load measures the recent load on the athlete and the chronic load measures the load on the athlete over a longer time period. Ideally you want to keep your ACWR in the 0.8-1.3 zone - the green zone - for maximum training improvement while making sure you don’t overtrain and incur injury.
Credit Card Process Automation
Skills: Java, Spring, SQL, Multi-threading, Automation, Web services, Data analysis, Data transformation, Report generation, SQL, Stored procedures, Cross-functional collaboration.
Senior Software Engineer at Newgen
-
Developed a multi-threaded scheduler using Java, Spring, and SQL to automate the credit card process, significantly reducing manual effort, and increasing efficiency.
-
Optimized complex SQL queries, and stored procedures, and performed data analysis and transformation to generate reports for regular audits.
-
Developed and maintained web services that integrated with existing systems and automated business processes, resulting in faster turnaround times and reduced errors.
-
Implemented security measures, such as encryption and tokenization, to protect sensitive user data, ensuring the system met industry security and data privacy standards.
-
Collaborated with cross-functional teams to align the product roadmap with business goals, successfully delivering a robust, scalable, and highly automated credit card processing system.
-
Conducted code reviews and mentored junior developers, fostering a culture of continuous learning and improvement within the team.
PowerBI Dashboards
Skills: Data visualization, Report design, Dashboard design, Custom visuals, DAX, Power Query, Power Pivot.
Personal Projects
University Projects
Skills: Cloud Computing, Ansible, Data Collection, CouchDB, Sentiment Analysis, Machine Learning, Google BigQuery, Feature Engineering, Natural Language Processing, Deep Learning.
Cloud Computing: Sentiment Analysis of Top 5 Premiers in Australia
​
For this project, I conducted a sentiment analysis of tweets related to the top 5 premiers in Australia. The aim was to gain insights into public opinion and attitudes towards these political figures. To achieve this, I utilized UniMelb Cloud Platform, Ansible, Twint, and CouchDB for data collection, storage, and analysis. The project involved designing and implementing a cloud computing infrastructure to collect and store large amounts of data from Twitter. I then utilized natural language processing techniques to analyze the sentiment of the tweets and generate insights. The project helped me develop my skills in cloud computing, data collection and storage, as well as natural language processing for sentiment analysis.
Machine Learning in Health: Predict Cardiac Arrest with Opioid Overdose in Patients
​
In this project, the main objective was to predict cardiac arrest in patients who had overdosed on opioids by implementing machine learning techniques. To achieve this, Google BigQuery was used to efficiently analyze and extract insights from the MIMIC-IV database. Furthermore, feature engineering and natural language processing techniques were implemented to preprocess and extract relevant information from unstructured data such as doctors' notes and discharge summaries. The machine learning model was trained and evaluated using a variety of techniques such as cross-validation and hyperparameter tuning to achieve optimal performance. Overall, the project demonstrated the potential of machine learning in healthcare for predicting and preventing life-threatening events.
Deep Learning: Human Action Recognition
This project focused on the application of deep learning techniques for human action recognition. Specifically, we developed and trained machine learning models, including convolutional neural networks (CNNs) and recurrent neural networks (RNNs), to classify various types of human actions. To achieve the best model performance, we employed optimization techniques such as hyperparameter tuning, cross-validation, and ensemble methods. The project aimed to improve the accuracy of human action recognition, which has many practical applications, including in video surveillance, robotics, and gaming.
Natural Language Processing: Rumour Detection and Analysis of tweets
​
​In this project, Natural Language Processing techniques were employed to detect and analyze rumors in tweets. The data was preprocessed using techniques such as tokenization, stemming, and lemmatization. The BERT model was used to develop and train the NLP model, and its performance was evaluated using metrics such as accuracy, precision, recall, and F1 score. The project aimed to develop a reliable and accurate system that can automatically identify and classify rumors on social media platforms, which can be used to mitigate the spread of misinformation.