Portfolio Exhibit
CG Texture Upscaler
Independent
Summary: I developed this Python-based feature-rich image upscaler that uses an ESRGAN architecture that has been trained on over 5000 computer graphics textures used in animations, XR and video games. The applicaiton offers advantages over its counter parts such as Topaz Photo AI and Hitpaw Photo Enhancer in that it is able to preserve the image channel details when upscaling as demonstrated here.
​
It provides a single model for use, namely the CG/art model that results in smoother, less noisy upscales, but offers the option to add noise from the original lanczos-upscaled image. It also supports exporting various image formats and various compression algorithms supported by wand, a Python binding of Image Magick, and offers numerous quality fo life features that make batch processing and sifting through images easier. Finally, the application comes packed with a CLI that supports everything that the GUI supports. The CG Texture Upscaler is a great solution for those working with computer graphics textures of all kinds based on the feedback of professional acquaintances of mines that have tested it.
​
The application is in the pre-release phase and is stable for use and can be downloaded here. It will remain in this phase (with various 0.x.x verions pre-releases) untill the full suite of unit and integration tests have been written.
​
Prominent technologies used: Python, PyTorch, OpenCV.
​
BERT/BART Review Sentiment Analyzer
Independent
Summary: I developed this Python-based review sentiment analyzer that features a BERT preprocessor and encoder, as well as a DNN classifier trained on over 1 million public user comments that I mined and their respective ratings to predict anonymized reviewer sentiment.
​
Prominent technologies used: Python, TensorFlow and TF Hub, Pandas/NumPy, Django, FastAPI, Docker.
TFRS Book Recommender System
Independent
Summary: I developed this Python-based TFRS recommender system application that accessible through it's unique API and that features an embedding network trained on over 70 000 books and over 700 000 public anonymized user data that I mined.
​
Prominent technologies used: Python, TensorFlow Recommenders, Pandas/NumPy, Django, FastAPI, Docker.
Toronto Police MCI Crime Location Prediction
Independent
Summary: I developed this Python-based crime location prediction application based on the MCI dataset provided by the Toronto Police Service. It's accessible through its unique API.
​
Prominent technologies used: Python, scikit-learn, Pandas/NumPy/PySpark, FastAPI, Docker.
Pairwise and Listwise TensorFlow Recommenders Book Rankers
Independent
Summary: I developed this Python-based recommender model based on the two tower recommender architecture to rank books to users using TensorFlow Recommenders 'Ranking' api.
​​
Prominent technologies used: Python, TensorFlow, Pandas/NumPy
Attention Seq-to-Seq English-Arabic Translation Machine
Independent
Summary: I developed this Python-based translation machine application that uses attention seq-to-seq RNNs to handle sequence Arabic-English NLP data which presented unique challenges due to the syntactic difference between the language families. It is accessible through its unique API.
​
Prominent technologies used: Python, TensorFlow (Keras), Pandas/NumPy, FastAPI, Docker.