HPA Dashboard is a reusable software package for imaging-based machine learning projects. It was initially developed for Human Protein Atlas microscopy data, but the workflow is designed to be adapted to other imaging datasets and downstream tasks. The package provides an end-to-end dashboard for exploring image data, analyzing class imbalance and label co-occurrence, monitoring model training, running inference, reviewing uncertain samples, and translating model outputs into practical next steps. A key feature is the built-in auto-research training agent. Instead of requiring users to manually monitor every training run in a notebook, the agent tracks training progress, evaluates model behavior across epochs, summarizes useful signals, and supports decisions such as whether to continue training, stop early, or adjust the next experiment. The goal of this package is to reduce repetitive setup in imaging projects. Rather than starting from a blank Jupyter notebook for every new dataset, users can begin with a working workflow and adapt it to their own research problem. Project repository: https://github.com/aimed-lab/HPA