Computer Science @ Drexel University · Data Science Minor · Expected December 2028
Software developer focused on machine learning, computer vision, and LLM systems. Currently building autonomous drone navigation at Exo Dynamics and leading a first-author ICML submission benchmarking LLM constraint-planning at Drexel. Previously a Machine Learning Engineer Co-op at Lockheed Martin, working on real-time aerial detection systems.
Interests outside of work: bodybuilding, making music in FL Studio, and poker.
Software Developer — Exo Dynamics Co. (Jan 2026 – Present)
Autonomous landing systems for drones using 3D pose estimation, YOLO obstacle avoidance (180K+ images, ~88% val accuracy), Kalman filtering, and optimized OpenCV/Python vision pipelines.
Machine Learning Engineer Co-op — Lockheed Martin (Oct 2025 – Mar 2026)
Real-time detection transformer on 40K+ drone images (~45% mAP), motion-compensated background subtraction, 2D Kalman filtering for bounding-box tracking, and a YOLO-Pose ship deck landing pipeline (≤1° rotational error over 100K+ synthetic renders).
LLM Student Researcher — Drexel University (Jan 2025 – Present)
First-author ICML submission benchmarking 6+ SOTA LLMs across 500+ constraint-planning tasks. Built a 3-stage async inference/evaluation/LLM-judge pipeline and a multi-pass constraint-feedback refinement loop achieving 2–4× accuracy gains.
Languages — Python, C++, SQL, Bash, TypeScript, JavaScript
ML & Deep Learning — PyTorch, TensorRT, YOLO, Vision/Detection Transformers, Scikit-learn, OpenCV
LLM & NLP — Hugging Face Transformers, LangChain/LangGraph, RAG, Fine-Tuning
Infrastructure — FastAPI, Docker, Linux, Git/GitHub/GitLab, CI/CD, REST APIs, NumPy, Pandas
PyTorch LayerNorm C++ Extension
High-performance LayerNorm in C++ (LibTorch) with full autograd via Pybind11. 22–58% CPU speedup, 52% Apple GPU improvement, output error <0.01%.
LLM Planning Research
Evaluation framework benchmarking SOTA LLMs (DeepSeek-R1, GPT-4o, etc.) on formal constraint-planning. Uncovered up to 90% hard-coded solutions in DeepSeek-R1 outputs. ICML submission.
OncoDeepMind
End-to-end PyTorch pipeline predicting cancer drug response from GDSC genomic data. 32% baseline improvement, deployed with FastAPI for real-time inference.
DragonFlow
Full-stack course scheduling platform for Drexel students. Gradient Boosting model (~90% val accuracy) with Flask inference API, built as Product Owner for a 4-member team.
Asteroid Trajectory Predictor
ML model trained on NASA asteroid data to predict impact probability for potential threats.
Heart Failure Predictor
Interactive web app with a custom-trained model predicting heart failure probability.
Releasing soon...
- Generative AI with Large Language Models — DeepLearning.AI & AWS (Coursera) · Aug 2024
- Launching into Machine Learning — Google Cloud (Coursera) · Apr 2025