Implemented and improved the Q-learning algorithm's decision-making process in complex, risk-laden environments
Authored a paper for SPIE titled "Modeling Adversary Behavior Using Q-Learning"
Summer 2023: D'Crypt AI Intern @ D'Crypt
Objective: Spearhead the development of a secure Chat-GPT alternative for code generation and debugging that mitigates the risk of leaking confidential company information
Conducted comprehensive research and rigorous testing of top Large Language Models (LLMs) for code generation locally and via cloud GPUs
Achieved a notable improvement of 14% in engineer coding efficiency
Demonstrated expertise in Python by designing engaging labs and homework assignments
Mentored students during class discussions, fostering a collaborative learning environment and providing valuable insights into the world of data science
Summer 2021: Birdeye (Virtual) Software Development Intern @ Birdeye
Employs graph theory to model adversary moves within a strategic environment
An improved Q-learning algorithm (Risky Q-learning) is implemented that takes into account rewards, risks, and different consequences to those risks associated with each move in the graph and chooses the most optimal path
An algorithm is implemented to negatively impact adversary decision making by modifying risk values in the decision making graph to reduce the adversary's average reward and success rate of their decisions to reach an end goal
Implemented Bezier curves and surfaces using de Casteljau subdivision, smoothing of meshes with averaged normals, loop subdivision for upsampling meshes
