I am a PhD student in Health (Health Systems Management & Health Data Analytics) at York University's School of Health Policy and Management, supported by the Connected Minds Doctoral Scholarship, and a member of the Health Equity and AI (HEAL) Lab, which uses AI to reduce health disparities and build fairness into health prediction models. Alongside my research, I work as a software engineer on health and education technology: a domain-specific language and VS Code tooling for patient care plans at Aurora Constellations, the ARCAND Map research network platform at the Dr. Gilles Arcand Centre for Health Equity, and PathwAI, an AI platform for credit transfer between Ontario institutions. I am based in the Greater Toronto Area.

Research Interests

• Neurosymbolic AI
• Knowledge Graphs
• Human-AI Collaboration
• Health Informatics
• Electronic Health Records
• Clinical Decision Support
• Computer-Aided Design for Healthcare
• Domain-Specific Languages

1. Medical Image Captioning on Chest X-Rays

   Skills: Python, Long Short-Term Memory (LSTM), CheXNET Model

   Extracted the information from Indiana University's dataset, consisting of almost 7500 images and 4000 reports.
   Organized the unbalanced by performing data pre-processing tasks like up-sampling and down-sampling.
   Captured the semantic meaning & relationship between the words by using GloVe vectors with 300 dimensions.
   Fine-tuned the CheXNET Model for encoding and achieved around 90% accuracy and a loss of less than 0.01.
   

2. Video Game Sales Dashboard

   Skills: Data Cleaning, Tableau

   Worked on the Video Game Sales dataset from Kaggle which includes 10 attributes and more than 16000 samples.
   Devised an interactive Dashboard on Tableau to find the insights and relationship between the genre, year and sales.
   

3. Currency Recognizer

   Skills: Research, Computer Vision, Scale-Invariant Feature Transform (SIFT), Python, MobileNet, Data Collection

   Collaborated on a project to replicate and enhance a research paper.
   Developed a fine-grained currency recognizer for the visually impaired using CONGAS and SIFT-based features.
   Engineered a system with exceptional precision and recall, employing a lightweight coarse classifier and fine-grained recognition.
   Led the creation of a diverse dataset (300 images per US banknote) with varied conditions for robust model training and evaluation.
   Implemented a pre-filtering classifier and utilized SIFT features to optimize recognition accuracy for visually impaired users.
   

4. Qualitative Bankruptcy Classification

   Skills: Analysis, Python, Multi Layer Perceptron

   Developed a Multilayer Perceptron (MLP) from scratch, incorporating Backpropagation and Stochastic Gradient Descent with Momentum (SDGM) algorithms.
   Attained a successful learning curve, demonstrating effective model training and optimization.
   Utilized parameters referenced from the paper 'The Discovery of Experts' Decision Rules from Qualitative Bankruptcy Data Using Genetic Algorithms' by Myoung-Jong Kim and Ingoo Han for an existing dataset.
   

5. Half Adder Circuit Design using Perceptron

   Skills: Logic Gates, Python, Feed Forward Perceptron

   Designed a Half Adder circuit using a Feedforward Perceptron, showcasing practical implementation skills in neural network architectures.
   Employed the Unit Step function as an Activation Function within the circuit design, demonstrating familiarity with various activation functions.
   Developed AND and NOT logic gates, subsequently constructing a NAND Gate, highlighting expertise in logic gate implementations and circuit design.
   

6. Online Vehicle Renting System

   Skills: Bootstrap, Django Framework, Python, Jinja3, SQLite3

   Designed and developed a web system using Django Framework for online vehicle booking/renting.
   Accelerated the performance with optimized programming & reduced system failure metrics for parallel requests.
   Fashioned the user interface using Bootstraps and CSS for easy single-button click navigation.
   Operated and loaded the live user, vehicle and rent request data using Jinja.