Computer Science Projects
A selection of software projects spanning interactive graphics, full-stack web apps, and systems programming
3D Interactive Airport Scene (Three.js)
- Engineered a fully interactive 3D airport environment from scratch using Three.js / WebGL — modeling a complete aircraft (fuselage, wings, engines, landing gear), hangar, fuel truck, baggage carts, and light poles with real-time cast and received shadows.
- Architected a dual-camera system with seamless switching between orbit overview and WASD first-person navigation; layered in raycasting for click-driven interactions — toggling individual light poles and animating the hangar door open/close.
- Designed a production-grade multi-light pipeline (directional, point, and spot lights with 2048×2048 shadow maps), a looping fuel truck animation, sine-wave beacon lighting, atmospheric fog, and tarmac/grass texture mapping for visual depth.
Neural Network from Scratch — MNIST Digit Classifier
- Engineered a complete two-layer feedforward neural network using NumPy only — deliberately avoiding ML frameworks to hand-derive and deeply understand every forward pass, gradient, and weight update from first principles.
- Designed a 784 → 10 (ReLU) → 10 (Softmax) architecture and trained it on 42,000 MNIST digit images, holding out 1,000 samples for unbiased evaluation; manually implemented backpropagation, one-hot encoding, and gradient descent.
- Achieved ~89% training accuracy in 500 iterations at α = 0.1, validating correctness of every numerical derivative and confirming the network learned meaningful digit representations from scratch.
HFT Strategy Backtesting Engine
- Engineered a Python backtesting engine for simulating high-frequency trading strategies combining RSI momentum signals, Bollinger Band mean reversion, and configurable stop-loss/take-profit logic.
- Ingested and processed millisecond-resolution tick data from CSVs, computing rolling statistics and trade signals across thousands of data points using Pandas.
- Delivered quantitative performance insights via Sharpe ratio analysis, drawdown tracking, and parameter sweeps — visualized through multi-panel Matplotlib dashboards for rapid strategy evaluation.
ClassConnect – QR-Based Attendance Platform (In Progress)
- Architected a full-stack QR-based attendance platform enabling professors to generate dynamic, time-expiring QR codes for fraud-proof student check-ins — eliminating proxy attendance entirely.
- Built the frontend in JavaScript / HTML / CSS and powered the backend with Firebase Cloud Functions and Firestore, delivering sub-second session validation and real-time dashboard updates for professors.
- Reduced attendance logging to under 30 seconds per session with 100% submission accuracy, cutting manual entry time by 90% and giving instructors instant visibility into live class attendance.
Pharmaceutical Inventory System
- Architected a modular pharmaceutical inventory management system in Java, applying OOP principles — encapsulation, inheritance, and polymorphism — to cleanly separate stock, supplier, and transaction logic.
- Automated multi-step financial calculations and inventory updates, cutting input errors by 15% and boosting processing speed by 25% over the manual workflow it replaced.
- Delivered real-time stock tracking and restocking visibility, giving pharmacists instant insight into inventory levels and measurably streamlining day-to-day business operations.
Code Analysis Tool (CAT)
- Engineered a WPF desktop application in C# / .NET that statically analyzes Java codebases and surfaces actionable complexity metrics — LOC, class counts, inheritance depth, and conditional branching.
- Built a structured analysis pipeline that processes 10+ Java files in seconds, extracting and visualizing code quality indicators to drive informed refactoring decisions.
- Reduced manual code review time by 25% and improved analysis accuracy to 90%, replacing error-prone line-by-line inspection with automated, reproducible metric reporting.
