RSSAmplifier

Blog

John Sutor

John Sutor's personal website.

johnsutor.com โ†—RSS feed โ†—13 posts

Latest posts

๐Ÿค– SO-101 Nexus

An open-source toolkit for training and deploying learned policies on the SO-100/SO-101 robotic arms.

๐Ÿ“ Some GRPO Training Notes

Some basic notes for effectively training with GRPO.

๐Ÿค” LLM Reasoning Datasets from Scratch

Learn how to quickly and easily create LLMs reasoning, from scratch.

๐ŸŸ Aquaponics

An in-apartment aquaponics system for growing herbs.

๐ŸŽจ Emoji Painter

A Transformer-based model that paints using emojis. You can provide images or emojis to paint with, and it will attempt to recreate images using emojis.

๐Ÿ‘จโ€๐Ÿ’ป Leetcode Study Tool

This package provides a command-line tool for interracting with Leetcode to create flashcards for study, which can then be imported into Anki.

๐ŸŒ Dead Simple I18n Using Google Translate

I18n (Internationalization) has never been simpler to implement using Google Sheets and Python.

๐Ÿฆฆ Do-it-yourself Jupyter Notebook Autograder with Otter Grader and Google Cloud

With the expansion of SciTeens curriculum program, we decided to set out to create our own Jupyter Notebook autograder for our website.

๐Ÿ“ฑ Fordle (Fake Wordle)

A Fake Wordle clone that lets you input your stats.

๐Ÿ“ท Leopardi (Class-based Synthetic Data Generator)

A class-based library for generating 3D rendered synthetic data. This library allows for a myriad of control over the specifics of the rendering engine, including but not limited to lens field of view, rendering angle, and mode by which to load a background and/or 3D model to be rendered. This library is also fully extensible, and can be installed from PyPi.

๐Ÿ”€ Replacing pre-trained Pytorch model layers with custom layers

My contempt for Batch Normalization and love of Dropout layers led me to upgrade pre-trained convolutional neural networks in the simplest fashion possible.

๐Ÿ’ป Synthblend (Command-line Synthetic Data Generator)

A simple-to-use command-line tool for quickly working with synthetic data. With a directory of background images and another with object files and accompanying textures, you can quickly generate thousands of synthetic images to supplement any dataset. This library interfaces with the Blender bpy library.

๐Ÿงช SciTeens

The website for SciTeens Inc. Built using NextJS with Server Side Rendering, with a Firebase backend, Google Kubernetes hosting, Google Functions, and OAuth 2.0 for backend operations, and search powered by Algolia. Website is actively maintained.