Projects¶

Currently Working On¶

Working on developing an ML model to optimize viral replicase sequences for RNA therapeutics. We aim to predict mutations on the saRNA (self-amplifying RNA) sequences to achieve three objectives: reduce the immune response, increase RNA replication rate, and increase protein expression rate.

Extended a recently published paper (Forager) by adding metrics to evaluate plasticity loss on existing models. I ported some old environments into a JAX based version (called Foragax) which allowed our group to run parallel experiments across seeds and environments. In this new environment, I tested new architectures and developed a Transformer + PPO architecture in Flax NNX which outperformed baseline DQN and PPO models.
Recreated (Contrastive Abstraction) by developing an ML pipeline from scratch. I created the data collection, sampling logic, hopfield model and the contrastive learning models and was able to replicate the paper's results of temporal state representation.
Created a robotic manipulation system that combines instance segmentation with moment-based visual servoing to enable a Kinova Gen3 robot arm to pick and place geometric shapes. I use a custom-trained YOLO model to extract polygonal masks of tabletop objects, which are converted into Image Moments. For the robot control pipeline, I use a ROS-based control stack that communicates with the vision node via ZeroMQ.