At Erets Space, I develop computer vision pipelines to extract and identify satellites and orbital debris from challenging, low-SNR optical imagery. I implement and evaluate PyTorch CNN and DETR architectures, focusing on reducing false positives and improving model reliability.
My work contributed to a third-place finish in Phase 1 of the SPARK 2026 Challenge, featured at the CVPR AI4Space Workshop. In my projects, I built a zero-shot food identification pipeline with YOLOv26n, OpenAI CLIP, and FAISS, and implemented a Vision Transformer from scratch in PyTorch.
