Hi, my name is Guneev Dhillon
I'm a

NotiFlow is a full-stack system I built to aggregate Canvas assignments, announcements, and exam schedules into a single, reliable iCalendar feed for students. I designed backend services in Python and Flask to fetch, normalize, and reconcile data across multiple APIs and scraped university sources, handling inconsistent formats and missing data. The application is containerized with Docker and is actively being deployed to a Linux-based DigitalOcean server, with attention to environment configuration, service orchestration, and production runtime behavior. This project emphasized end-to-end ownership of a data pipeline, disciplined interface design, and debugging across service boundaries in a production-oriented setup.

Shipped a web app that turns prose into manga panels using OpenAI image APIs. Implemented scene extraction, prompt chaining, and panel layout; handled rate limits and content filters. Developed during UBC's ProductX Hackathon.

Neural Network Name Generator is a self-directed project where I implemented a character-level neural network from scratch to generate plausible names. Following Andrej Karpathy’s tutorial as a conceptual guide, I manually derived and coded the forward pass, backpropagation, and softmax-based training loop using NumPy to understand the underlying mathematics rather than relying on abstractions. I trained the model on name datasets, debugged gradient flow and convergence issues, and experimented with architectural and hyperparameter changes to improve output quality. After validating the core implementation, I reimplemented the model in PyTorch to compare performance, training stability, and development tradeoffs between low-level and framework-based approaches.

Transaction Analyzer is a Python-based tool I built to analyze personal financial statements by parsing CSV exports and normalizing vendor data. The system groups transactions by vendor and time period, computes credit and debit totals, and produces net summaries to surface spending patterns. I implemented filtering, sorting, and aggregation logic using pandas, along with one-click export for downstream use. This redcuses manual bookkeeping time by 90%.

Catcher Game (Brainrot Game) is a 2D Unity game I built in C# to explore core gameplay systems and engine-level workflows. I implemented an object spawner, real-time input handling, 2D physics interactions, and collision-based scoring, along with a UI for game state feedback. The project emphasizes clean C# scripts, reusable prefabs, and straightforward state management within Unity’s component model. This served as a focused exercise in building a complete, playable game loop rather than isolated mechanics, emphasizing modular design and good coding-practice to build a maintainable system.