Frederick Roman's Projects
Svelte binary-coded decimal clock.
full-stack web app that scores cars using neural network trained on UCI car evaluation dataset
Collatz conjecture graph calculator that calculates and displays in-browser the sequence of large numbers.
This project demonstrates how to use deep neural networks to generate stylized images given an input image and an input art style.
This is a divorce predictor using a neural network. The moded is trained on the UCI dataset using tensorflow, then converted to tfjs. (This is the model used in the Divorsee app.)
Private divorce predictor using tensorflow on a prerendered Angular progressive web app.
This is an etymology prediction RNN model that uses data scraped from Wiktionary.
Unofficial minified fastetext API. Use it to run NLP DL models that require word embeddings on the client-side.
Config files for my GitHub profile.
The client-side of Navlead. Navlead is an deep-learning assistant chatbot for navigation in 3D enviroments.
Nictographic cypher tool. It uses an extended version of the nyctography cypher system to encode regular text.
This notebook shows how to train a neural network to solve quadratic equations where the coefficients are on [0, 1)
Random English-sounding (phonotacticaly correct) word generator that runs on a full-stack web app with public API. (Vue + Node + Express)
Reading ease scoring web app. It calculates the US school reading level of a text client-side using neural networks.
Signature verifier. It uses DCT to extract the signature's features and then classifies them using SVM.
Neural network model that predicts the number of syllables in an English word. It shows its creation end-to-end: from data collection to evaluation of various models. One of the explored models is used in the Readgauge app.
Voxel-based 3D modeling online editor.
AI-powered zoo animal classifier. It uses a neural network trained on the UCI Zoo dataset with Brain.js. It is shown in a PWA React app.