Projects

• ResearchKG — Free research citation network tool: visualize connections in the literature and explore a field from a seed paper (DOI / title / concept). Desktop version coming soon. · demo
• AnkyloRL — MuJoCo + curriculum PPO for Ankylosaurus magniventris(the coolest dinosaur) to validate paleobiologically grounded behavior through sim.
• GNNs from scratch — NumPy-only implementations of graph neural network layers and models: a learning exercise in message passing without frameworks.
• GraphRL — Small RL experiments on graph-structured combinatorial problems such as PPO for graph coloring, Q-learning for TSP.
• OpenRECAP — Deep RL final project, open-source pipeline adapting RECAP (advantage-conditioned offline RL) to fine-tune SmolVLA for tabletop pick-and-place, with MuJoCo sim evaluation, LeRobot demo recording, BC pre-training, and autonomous rollout.