Open-source courses, AI systems, ML experiments, and developer tools. Each project links out to the source, while this page gives recruiters and collaborators the shape of the work quickly.
AI-assisted paper study for researchers and builders.
A solo-built SaaS product for searching, studying, summarizing, and chatting with research papers. It combines agentic search, recommendations, paper chat, background jobs, Postgres, and custom ML deployment across a large Arxiv-scale corpus.
Enterprise-scale simulation of social media behavior with AI agents.
A multi-agent LLM simulation system where AI personas read news, form opinions, post content, and influence each other inside a sandbox. Built for consumer-behavior research and enterprise strategy at SocialTrait.
Client-side text generation for call-center agent autocomplete.
A production ML system for assisting live agents as they type, built before ChatGPT-era tooling became standard. The work involved lightweight autoregressive models, conversational embeddings, latency constraints, deployment, and user-facing product impact.
Recursive Language Models for long context understanding.
A research-oriented implementation of Recursive Language Models focused on breaking long-context tasks into structured recursive passes. The project packages the core idea with runnable examples, a quickstart flow, and a terminal interface for experimentation.
Lightweight neural text utilities and experiments.
A small collection of neural text utilities and experiments for practical NLP work. This is closer to a toolbox repo: lightweight scripts, reusable patterns, and text-processing experiments that support larger ML projects.
Open-source Context Engineering course with DSPy.
A hands-on course repository that teaches context engineering by building DSPy modules level by level. It turns prompting, retrieval, routing, evaluation, and optimization into practical notebooks instead of abstract advice.
Agentic Memory architecture from scratch in DSPy.
A from-scratch memory layer for LLM agents inspired by Mem0, implemented with DSPy. It covers how memories are created, searched, updated, and injected back into agent workflows so conversations become stateful over time.
Converting text-only language models into Vision Language Models.
A compact VLM training project showing how image encoders, projection layers, and text-only language models can be connected into a multimodal system. The repo includes architecture diagrams and output examples to make the training path concrete.
Universal synthetic dataset generation from raw data.
A data augmentation toolkit for text tasks that helps create synthetic training examples from raw data. The project focuses on practical dataset expansion patterns for NLP workflows where labeled data is limited.
Multi-agent reinforcement learning obstacle avoidance with CTDE.
A multi-agent reinforcement learning environment and training setup for navigation and obstacle avoidance. It demonstrates MAPPO-style centralized training with decentralized execution across several navigation scenarios.
DFC-VAE: A Variational Autoencoder for Dynamic Feature Composition.
A Variational Autoencoder for Deep Feature Consistent Images. Manipulate the latent space directly by training a VAE, and then use arithmetic operations directly in the latent space.
Morphology-driven coordinated control for articulated characters.
A Clemson research project on physics-based character control. The system learns coordinated locomotion from latent coactivation structure, producing control policies for articulated characters such as cheetah-like agents.
Training physics-based characters to dance to music.
A graduate research project from the MS thesis stream at Clemson. It explores deep reinforcement learning for expressive, physics-based character motion, connecting locomotion control with music-conditioned dance behavior.
Terminal-native notebook tooling for animated technical walkthroughs.
A terminal-native notebook rendering tool for creating animated technical walkthroughs. It is built for educational videos where code, highlights, callouts, and notebook state need to render cleanly in sequence.
Network-science project modeling movie data as a hypergraph.
A graph-learning exploration where movies, people, genres, and relationships are modeled as higher-order structures instead of simple pairwise edges. The project sits at the intersection of network science, recommender-style representation, and graph analytics.
Biomedical ML project for autism analysis from fMRI connectivity.
A biomedical data science project exploring functional brain connectivity patterns for autism classification. It applies graph/network-style representations to fMRI-derived signals and focuses on clinically meaningful structure rather than raw black-box prediction alone.
Self-play reinforcement learning experiments in Unity.
A Unity-based self-play project for training agents through competitive or adversarial interaction. The focus is on emergent policy improvement from agents learning against evolving versions of each other.