The ASL‑Kimpinde Set is a small, custom dataset collected under controlled conditions by Samuel L. Kimpinde. It contains 250 video samples across 50 isolated ASL word classes, with 5 samples per class. Recordings were made using a Lenovo V15 G2 laptop webcam at 1280×720 resolution, 30 FPS. Metadata logs include filename, class, environment, and recording parameters. Capture scripts and a README file are provided to ensure reproducibility. The dataset was designed to augment underrepresented classes in the WLASL dataset and to demonstrate domain‑aware quantization calibration for TinyML sign language recognition systems.