Perceptilabs

The product is a visual, dataflow-driven API for TensorFlow that helps users build, tune, and understand deep learning models through interactive components and real-time visualizations instead of writing low-level code manually. It is for developers and data teams in businesses that build machine learning models and need transparency in model development. It is positioned as a visual alternative to code-only TensorFlow development that auto-generates editable code and surfaces model behavior visually.

Key features

• Visual dataflow API for TensorFlow
• Drag-and-connect components for model building
• Component settings tuning
• Auto-generated output visualizations
• Real-time visualization updates
• Auto-generated editable TensorFlow code
• Automatic input/output shape calculation
• Suggested configs and hyperparameters
• Model architecture visualization
• Tips, warnings and errors while building
• Debug model and code with instant feedback
• Real-time granular training visualizations
• Model performance and predictions overview
• Visualize weights, biases, gradients, outputs
• Detect vanishing or exploding gradients