Nahtadi
Prayer times and Qibla for iOS. Privacy-first and fully offline, built in Swift and SwiftUI.
PROJECTS
Everything I've built.
iOS apps, ML models, GPU kernels, and the tools in between. Case studies where there is a real story to tell. Straight to the code everywhere else.
14 of 14 projects
Brent's Method on CUDA
The first CUDA implementation of Brent's root-finding method. One solver per GPU thread with bit-identical fp64 results.
Radar Plotting Trainer
A maneuvering board that grades your plot, from two radar observations to the maneuver that opens the CPA.
CG Pilot Training Tracker
A training-data report adopted at every Coast Guard air station. Over a week of compiling became three minutes.
On-Device LLM Summarizer
A 1.5B model fine-tuned, quantized to 4-bit, and running entirely on an iPhone Apple ruled out for on-device AI.
NewGame+
A hybrid Python and Prolog recommender that picks your next game from the ones you already love.
GitHub (opens in a new tab)Logic programming
Mini Compiler
A table-driven LL(1) compiler that turns Pascal-like source into executable Python. 78 tests, zero dependencies.
GitHub (opens in a new tab)78 tests · 71% cov
Prayer-Time Algorithm Library
The zero-dependency Python library behind Nahtadi. Astronomical algorithms with more than eight calculation methods.
GitHub (opens in a new tab)105 tests · 90%+ cov
Cycloidal Drive Creator
Generates parametric equations for cycloidal drive rotors that import straight into SolidWorks.
GitHub (opens in a new tab)28★ · MIT
Image Watermark Remover
A Pix2Pix GAN in PyTorch that removes watermarks, trained on roughly 16,700 images.
CG Helicopter Inventory
An inventory system for helicopter maintenance, in production in San Diego. Part searches are 85 percent faster.
USCG · PRIVATE−85% search time
ASL Letter Detector
Real-time ASL letter detection with a retrained YOLOv5 model. A practice tool for students.
GitHub (opens in a new tab)Computer vision
Wildfire Likelihood Predictor
A TensorFlow model that predicts California wildfire likelihood from weather and historical data.
GitHub (opens in a new tab)ML classification
Reddit NLP Classifier
Which subreddit said it? Naive Bayes over CountVectorizer at 76.8 percent test accuracy.
GitHub (opens in a new tab)76.8% accuracy