WORK

Everything, in one place.

Projects, research, and notes — filter by kind to narrow the list.

01Edge AI

YOLO Edge Optimization

Took a YOLO detection model from research checkpoint to deployment across three accelerators and four runtimes, measuring what every optimization actually costs in latency, accuracy, and size. Found that export is 3.50× faster on NVIDIA TensorRT and 1.69× slower on Apple Core ML for the identical graph, and traced the regression to ten tensor-indexing operators in the NMS-free detection head that split the Core ML graph into eight partitions instead of five. Naive INT8 quantization collapsed the head to 0.000 mAP across all 5,000 COCO validation images without raising a single error; excluding 95 head nodes recovered it, and TensorRT quantized the same head successfully, establishing the failure as a quantizer limitation rather than an architectural one. Six findings documented alongside five hypotheses that failed under testing.

PyTorchYOLOONNX RuntimeCoreML

View project

02Computer Vision

Structured Ablation of YOLOv8-L on VisDrone

Four modifications from the UAV small-object detection literature — attention, a stride-4 detection head, PIoU regression reweighting, and inference-time tiling — applied to one YOLOv8-L baseline under a single protocol. None beat the unmodified model. The most-recommended fix cost 0.148 mAP50 and reduced small-object recall by 31%.

PyTorchYOLOv8UltralyticsVisDrone2019-DET

View project

03RAG / LLM Evaluation

RAG Evaluation Harness

Seven retrieval configurations benchmarked against a hand-verified eval set over a corpus of computer vision papers. No configuration beat plain dense retrieval on recall@1 — every added component improved recall@5 while making the top result worse. 14 findings, 10 of them negative.

PythonChromaDBsentence-transformersBM25

View project

04Medical AI

Swin Transformer for Histopathologic Cancer Detection

A Swin-Tiny Vision Transformer fine-tuned for binary tumor classification on PatchCamelyon histopathology patches, reaching ROC-AUC 0.955 and PR-AUC 0.99 at 85.7% accuracy. Evaluated on sensitivity and specificity rather than accuracy alone, because the false-negative cost in a diagnostic setting is not symmetric.

PyTorchSwin TransformerVision Transformerstimm

View project

05Representation Learning

Supervised Transfer vs. Contrastive Pretraining in GI Endoscopy

A controlled comparison of supervised transfer learning against SimCLR contrastive pretraining for gastrointestinal endoscopy classification, across ResNet50, EfficientNet-B0, and ViT-B/16. SimCLR reached 56.72% accuracy where supervised transfer reached 89.19% — and the vision transformer lost to both CNNs.

PyTorchSimCLRSelf-Supervised LearningResNet50

View project