3D cell culture and more recently Organ-on-Chip formats can accurately recapitulate many aspects of in vivo cell function, protein expression and metabolism, that diverge between conventional adherent cell culture on surfaces and actual tissue. Despite these benefits, adherent (2D) cell culture remains the predominant cell culture format for cell based high-throughput assays in drug development. The large numbers of cells required for 3D spheroids and Organs-on-Chips, the limited scalability of these large 3D formats to tens of thousands of conditions, and automation of tissue analysis are key challenges hindering widespread use of 3D models in early drug development. The cost effective use of 3D models in early drug development will require new minimal 3D cell models to be developed, produced and handled at the scale of millions of spheroids and finally assayed and analyzed rapidly. In this presentation, we address the production of spheroids by the hundreds of thousands using an automated droplet microfluidics based platform for high throughput production of millions of miniaturized spheroids. Automation of tissue analysis presents an additional challenge. Deep learning approaches, such as the use of convolutional neural networks have proven useful in automating tasks such as image classification and segmentation, previously out of reach to traditional machine learning. We present the training of a CNN model for automated spheroid classification, and apply it to optimize and improve our understanding of the process of cell aggregation and spheroid formation in microfluidic droplets. We apply our CNN model to the classification of mini-spheroid formation from 3 different cell lines over time and across a number of different droplet formulations to map out incubation time and physico-chemical environment requirements for mini-spheroid formation. We are currently extending the use of our CNN model to classify responses to spheroid assembly environmental conditions and a larger range of cell lines and primary cells.