▶

CollaBand — View Demo↗

CollaBand is a full-stack web application that enables remote creators to collaboratively edit music through live, joinable sessions. The platform includes a custom music editor, interactive user dashboards, and a versioned changelog that records project updates and edit history.

As the Lead Front-End Engineer and UI/UX Designer on a four-person capstone team, I designed and built the entire client-side experience. I handled the full product design process, including branding, layout, and user flow, and developed the responsive ReactJS interface. I orchestrated the real-time data layer by integrating client-side WebSockets with our Django and PostgreSQL backend, ensuring high-concurrency state management and low-latency synchronization across all active sessions.

Tech Stack: ReactJS, Django, Python, PostgreSQL, PubNub (WebSockets), JavaScript, HTML, CSS, UI/UX Design

View Technical Report [PDF] ↗

In this project, I engineered a data-driven analytical framework to quantify and predict the "Performance Gap" between manufacturer-provided laboratory fuel ratings and actual real-world vehicle fuel consumption. Utilizing a telemetry dataset of 600 trip-level records spanning 25 distinct vehicle models and multiple fuel configurations, I developed a comprehensive machine learning pipeline. This involved performing exploratory data analysis to evaluate operational variables and architecting a three-tier predictive modeling framework to accurately forecast true fuel efficiency under unpredictable, real-world driving conditions.

Acting as the lead data analyst and modeler, I successfully built a predictive solution that empowers fleet managers to forecast fuel requirements with high precision, enabling a strategic shift toward optimized, data-driven vehicle procurement. I was able to run Multiple Linear Regression, Gradient Boosting, and Random Forest models, conducting feature importance analysis to reveal that manufacturer-rated base mileage predominantly drives real-world performance, significantly outweighing other dynamic operational factors. This transition from theoretical laboratory estimates to a data-driven reality allows organizations to efficiently allocate vehicle resources while minimizing the financial risks of performance volatility.

Tech Stack: Python, Pandas, Scikit-Learn (Multiple Linear Regression, Gradient Boosting, Random Forest Regressors), Seaborn, Matplotlib, and Jupyter Notebook.

View Technical Report [PDF] ↗

In this project, I developed a discrete‑event simulation model to analyze and optimize the operational efficiency of a banking system. Using a dataset of historical customer service records, I performed statistical analysis to identify the appropriate interarrival and service time distributions. I then built a digital twin of the bank’s routing and service network, including Sign‑In, ATM, and Teller stations, to evaluate baseline performance and quantify the operational gap created by unpredictable real‑world customer traffic.

Acting as the lead simulation modeler and data analyst, I successfully scaled the environment to test various resource configurations under heavy operational loads. I was able to run baseline models alongside automated hardware expansion and augmented human staffing scenarios to pinpoint the true service bottlenecks within the branch's infrastructure. By evaluating the system's performance metrics across these models, I established that investing in human service capacity significantly outperformed adding more machines, providing a data-backed roadmap to reduce physical crowding and improve total customer processing times.

Tech Stack: Simio, Microsoft Excel (PivotGrid), Discrete-Event Simulation (DES), Queueing Theory, Statistical Distribution Fitting.

View Technical Report [PDF] ↗

I engineered a fault-tolerant Web-to-PDF automation script using Python and Playwright to completely eliminate manual data entry. Designed to operate autonomously in highly volatile web environments, the script mimics a returning human user to evade aggressive bot detection and physically extracts rendered DOM elements. By integrating automated byte-size validation, algorithmic natural sorting, and strict local storage management to prevent memory bloat, this automation seamlessly sanitizes and compiles perfectly formatted PDFs issue by issue without human intervention.

As the sole developer for this independent project, I designed the entire end-to-end architecture. I handled the complete extraction lifecycle by implementing playwright-stealth to strip automation flags, configuring persistent Chromium contexts via local user_data_dir routing, and building interactive CAPTCHA fallback loops.

Tech Stack: Python, Playwright, playwright-stealth, Chromium, img2pdf, natsort, Regex , URL Parsing (urllib)

View on GitHub