David W. Ludwig II, PhD

Portfolio

The following are my featured projects that I have worked on over the years.

Featured Projects

• SetBERT

  A transformer-based deep learning platform for contextualized embeddings and explainable predictions from high-throughput sequencing data. Published in Bioinformatics (2025).

  Tech: Python, PyTorch, PyTorch Lightning, NumPy, Pandas, Matplotlib, TensorFlow/Keras, Weights & Biases, Slurm, dnadb
  Role: Lead developer and first-author
  Status: Published
  Highlights:

  • Developed the first foundation model architecture for high-throughput sequencing (HTS), leveraging a set-based transformer to capture interactions between thousands of short-read DNA sequences
  • Built a custom high-throughput sequencing data container format (dnadb) for efficient on-disk random access during large-scale training
  • Trained models in a distributed fashion across 5 compute nodes using Slurm, with coordinated multi-GPU pipelines for large-scale pretraining
  • Pretrained on ~280,000 16S rRNA samples from 970 studies obtained from the QIITA platform
  • Achieved state-of-the-art genus-level classification accuracy (95%), outperforming existing single-sequence models
  • Integrated a full explainability pipeline via attention attribution, identifying biologically relevant taxa that drive model predictions
  • Deployed on Hugging Face and via a plugin (q2-deepdna) for the QIIME 2 bioinformatics platform

• Logisim Emulator

  A programmatic emulator for running Logisim circuit files, designed as a backend library for an automated grading system.

  Tech: TypeScript, Node.js, Mocha, Chai, NYC, Coveralls, Travis CI
  Role: Solo developer
  Status: Open Source (Published on NPM)
  Highlights:

  • Achieved 95% unit test coverage with continuous integration and automated reporting
  • Emulates digital logic circuits assembled through Logisim for automated assessment

• Star Blaster

• A 2D Asteroids-inspired cooperative arcade game built with Pygame, where two players team up—one piloting a ship and the other firing at incoming asteroids of varying size and speed. Features custom graphics, original music, and full gamepad support. Awarded 2nd place in the Hacker’s Choice Award. See featured blog post.

  Tech: Python, Pygame
  Role: Team lead and co-developer
  Status: Hackathon project
  Highlights:

  • Led a team of 5, designing the architecture and orchestrating development during a 36-hour hackathon
  • Implemented the main game engine logic and core gameplay functionality
  • Integrated art from graphic design team

Research & Deep Learning Projects

• DNAGAST

  Transformer-based generative adversarial model for producing synthetic high-throughput sequencing data. Published in The International FLAIRS Conference Proceedings (2025).
  Tech: Python, PyTorch, NumPy
  Role: Lead researcher and publication first-author
  Status: Published
  Highlights:

  • Created the first generative model capable of producing full, realistic high-throughput sequencing samples

• Context Learning (cTL)

  Context learning model that incorporates environment feedback to dynamically infer current context in both supervised and reinforcement learning settings.
  Tech: Python, PyTorch
  Role: Lead researcher
  Status: Published
  Highlights:

  • Designed an architecture that dynamically infers task context without explicit context input through feedback
  • Integrated into a reinforcement learning framework with demonstrated multi-task adaptability

• AC-GAST

  Custom AC-GAN variant of Generative Adversarial Set Transformers based on Stelzner et al.
  Tech: Python, TensorFlow/Keras
  Role: Lead researcher
  Status: Research prototype Highlights:

  • Created a generative model that conditionally generates 2D point clouds resembling MNIST digits

Desktop Applications

• Logician

  A prototype logic circuit emulator featuring a custom themed Qt interface with touchscreen support. Includes a drag-and-drop side panel for adding logic gates, along with selection, box selection, rotation, and movement tools for circuit design.

  Tech: C++, Qt
  Role: Solo developer
  Status: Prototype (UI-focused)
  Highlights:

  • Designed and implemented a custom cross-platform Qt UI for logic circuit editing
  • Supports drag-and-drop, box selection, rotation, and touchscreen interaction
  • Built as a foundation for a future educational circuit simulation platform

• Intercraft Launcher

  A custom desktop game launcher built as a replacement for the default Minecraft launcher. Allows users to select and download specific Minecraft versions or launch the Intercraft modpack, with automatic version file management and account-based login. Features a dashboard displaying the user’s Minecraft skin, live server statistics via a REST API, and real-time player lists retrieved directly from the server.

  Tech: JavaScript, Node.js, Electron
  Role: Lead developer
  Status: Open Source
  Highlights:

  • Incorporated version and modpack management with user account login
  • Designed and implemented a real-time dashboard with skin rendering, server stats, and online player queries
  • Implemented RESTful API integration for server communication

• Volume Pump

  A lighthearted interactive desktop widget that controls system volume via a virtual air pump. Pumping inflates the widget to raise volume, while air gradually leaks out over time—requiring users to keep pumping to maintain volume.

  Tech: Electron, Node.js, JavaScript
  Role: Solo developer
  Status: Deployed
  Highlights:

  • Built a custom Electron desktop UI with real-time system volume integration

Websites

• jimmiegreid.com

  A full-stack single-page application built for a master carpenter based in Atlanta, GA, showcasing his portfolio of projects and press features.

  Tech: Nginx, Laravel, PHP, MySQL, Vue.js, JavaScript, Sass, HTML
  Role: Full-stack developer
  Status: Deployed
  Highlights:

  • Developed a responsive SPA architecture with dynamic content rendering via Vue.js
  • Designed MySQL-backed CMS functionality for managing projects and media
  • Integrated contact form with backend validation and mail routing

• intercraftmc.com

  A full-stack web platform for the Intercraft Minecraft server, featuring member showcases, mod information, live server metrics, and user authentication. Includes a RESTful API backend and periodic skin rendering jobs to display members’ Minecraft avatars.

  Tech: Apache, Laravel, PHP, MySQL, TypeScript, Sass, HTML
  Role: Full-stack developer
  Status: Deployed
  Highlights:

  • Built RESTful API for server statistics and member data
  • Implemented token-based authentication and application form system

Collaborative / Game Projects

• Multiplayer Minesweeper

  A real-time competitive multiplayer version of Minesweeper built during a 36-hour hackathon by a team of four developers. Supports multiple concurrent game sessions with up to 4 players each competing on a large shared board. Players are assigned colors, and cleared squares are marked on the board in their color for all participants—whoever clears the most tiles wins.

  Tech: TypeScript, Mustache, Sass, Webpack, Express, WebSockets
  Role: Lead developer (full-stack)
  Status: Hackathon project
  Highlights:

  • Orchestrated the team and project structure
  • Co-developed a real-time multiplayer game server and frontend in under 36 hours
  • Co-designed large shared game boards with real-time synchronization across clients

• OpenComputers-Reimagined

  A complete from-scratch rewrite of the legacy OpenComputers mod, designed to integrate seamlessly with the existing OpenComputers ecosystem while modernizing its architecture for current versions of Minecraft. Built as a multi-platform mod compatible with both NeoForge and Fabric via the Architectury framework, extending complex functionality on top of the large Minecraft codebase.

  Tech: Java, Gradle, NeoForge API, Fabric API, Architectury API
  Role: Lead developer
  Status: In development
  Highlights:

  • Leading development of a major reimplementation project, coordinating contributions and architecture decisions
  • Implemented cross-compatibility infrastructure to support both NeoForge and Fabric loaders
  • Built clean, modular architecture on top of the extensive Minecraft modding codebase

Libraries & Frameworks

Python

• dnadb

  High-performance container format for storing and streaming high-throughput sequencing data, built on a Lightning Memory-Mapped Database (LMDB) backend for efficient random on-disk access

  Tech: Python, LMDB, unittest
  Role: Solo developer
  Status: Open source (published on PyPI)
  Highlights:

  • Used as the primary data backend in SetBERT and DNAGAST

• deepctx

  A modular framework for rapidly creating configurable deep learning training and evaluation scripts. Framework-agnostic and designed for flexible experimentation, it automatically generates CLI arguments, sets up reproducibility features (random seeding, logging, W&B), and orchestrates training/evaluation routines with pluggable modules.

  Tech: Python, TensorFlow/Keras, PyTorch, Weights & Biases, unittest
  Role: Solo developer
  Status: Open source (published on PyPI)
  Highlights:

  • Designed and implemented a modular configuration system to enable flexible plug-and-play experiment components
  • Engineered automatic CLI generation, seeding, and logging infrastructure to standardize experiment setup
  • Integrated experiment tracking via Weights & Biases for real-time monitoring and reproducibility
  • Used this framework internally to develop and train multiple published deep learning models

• explainable-attention

  A library that patches existing PyTorch transformer architectures to extract attention scores and apply explainability techniques. Currently implements Self-Attention Attribution from Yaru Hao et al., enabling interpretable attention maps from pretrained models.

  Tech: Python, PyTorch, unittest
  Role: Solo developer
  Status: Open source (published on PyPI)
  Highlights:

  • Developed a model-patching framework to inject attention-tracking hooks into existing PyTorch transformer architectures
  • Implemented the Self-Attention Attribution explainability method for visualizing attention contributions
  • Built a unit-tested pipeline for extracting, processing, and visualizing attention scores from pretrained models
  • Packaged the tool for public use and distribution via PyPI

Node.js / TypeScript / JavaScript

• Commander

  A library for rapidly setting up command processing in Discord bots, with built-in support for privileges, middleware, and guards for fine-grained control over command execution.

  Tech: TypeScript, Node.js, Mocha, Chai
  Role: Solo developer
  Status: Open source
  Highlights:

  • Designed and implemented a modular command-processing framework with built-in role-based permission handling
  • Developed an extensible middleware and guard system for injecting custom pre- and post-command logic

• TVDB API Client

  A feature-complete REST API client for TheTVDB, fully typed with 90% unit test coverage and CI. Published on NPM.

  Tech: TypeScript, Node.js, Mocha, Chai
  Role: Solo developer
  Status: Open source (published on NPM)
  Highlights:

  • Engineered a fully typed client library for TheTVDB API v4 with comprehensive endpoint coverage
  • Built a 90% unit-tested codebase using Mocha and Chai, with automated coverage reporting
  • Set up a CI pipeline to run tests, build artifacts, and publish releases to NPM

• TMDB API Client

  A RESTful API client for The Movie Database (TMDB) with TypeScript typings and automated CI.

  Tech: TypeScript, Node.js, Mocha, Chai
  Role: Solo developer
  Status: Open source
  Highlights:

• Developed a typed wrapper for the TMDB API to simplify data access in Node.js environments
• Implemented automated unit tests and continuous integration for build reliability
• Created modular request handling and error management for stable API interactions

C / C++

• OpenSHMEM from Scratch

  A graduate-level semester-long project implementing the OpenSHMEM specification from scratch in C, using PMIx as the communication backbone.

  Tech: C, PMIx
  Role: Solo developer
  Status: Open source (academic project)
  Highlights:

  • Implemented core OpenSHMEM one-sided communication primitives directly against the PMIx backend
  • Designed memory management and synchronization mechanisms to coordinate data movement across nodes
  • Delivered a fully working standards-compliant prototype from the ground up

PHP / Web

• Pathfinder

  A Composer package that provides an HTTP routing framework for PHP, inspired by Laravel’s routing system with middleware and controllers.

  Tech: PHP
  Role: Solo developer
  Status: Open source
  Highlights:

  • Designed and implemented a lightweight routing engine with middleware and controller support
  • Built a clean, expressive API for defining and managing HTTP routes in PHP

• Forge

  A Composer package implementing a custom templating engine for PHP, with support for component slots, inheritance, conditional statements/loops, and formatted output. Includes a VS Code syntax highlighting extension.

  Tech: PHP
  Role: Solo developer
  Status: Open source
  Highlights:

  • Designed a clean and easy to use custom language syntax
  • Engineered a full-featured templating engine with custom lexer, parser, and transpiler from scratch
  • Implemented an automated caching system for fast and automatic rendering
  • Developed a companion VS Code language extension for syntax highlighting and developer tooling

• Scout

  A Composer package that provides a feature-rich form validation framework for PHP, including built-in constraints and automatic error message generation for form fields.

  Tech: PHP
  Role: Solo developer
  Status: Open source
  Highlights:

  • Built a modular validation engine with a rich library of built-in field constraints
  • Implemented automatic error message generation and field binding logic

Tools & Systems Applications

• q2-deepdna

  Plugin for the QIIME 2 bioinformatics platform for using pretrained SetBERT models.
  Tech: Python, QIIME 2
  Role: Solo developer
  Status: Open source

• Intercraft-SFTP

  Custom SSH/SFTP server written in Python with a virtual file system for sandboxed file transfers.

  Tech: Python
  Role: Solo developer Highlights:

  • Implemented a fully functional SSH/SFTP server from scratch, handling session management and protocol compliance
  • Integrated authentication directly with the Intercraft user database shared by the website and launcher
  • Designed and built a virtual file system layer to emulate file operations in an isolated sandboxed environment
  • Developed custom handlers for file transfer commands, directory navigation, and permission enforcement
  • Delivered a secure and extensible framework for controlled remote file access within the Intercraft ecosystem

Coding Challenges & Problem Sets

• Advent of Code Solutions

  Solved 100+ algorithmic puzzles from the annual Advent of Code challenge, covering topics like dynamic programming, graph theory, parsing, and combinatorics.

  Languages: Python, C++

• Kattis Solutions

  Solved 200+ competitive programming problems on the Kattis platform, focusing on algorithmic efficiency, data structures, and computational problem solving.

  Languages: Python, C++
  Highlights: #1-ranked competitor in the state of Tennessee