At Noèse, I built and fine-tuned an SLM-style NLP stack using MiniLM embeddings and an LLM to parse unstructured profiles and job postings. I diagnosed failure cases and validated changes against a production classifier with agreement metrics before integration.
At ZettaByte, I built an ML matching engine pairing student profiles with job descriptions for deployment across client universities. Iteration on retrieval and ranking against a held-out evaluation set improved match quality by 15%.
On my Exoplanet Direct Imaging Pipeline project, I built an end-to-end high-contrast imaging pipeline to recover faint planetary signals from low-SNR telescope images. I validated it with injection-recovery testing.
My computer vision work also includes a reproducible self-supervised learning benchmark, and an algorithmic trading project that placed 86th among 12,500+ teams. I work primarily in Python, including PyTorch, OpenCV, and scikit-learn.
