Show HN: BlockDL – A FOSS neural net sketchpad with shape checking and live code

Hi everyone. Designing neural network architectures is inherently a visual process. Every time I train a new model, I find myself sketching it out on paper before translating it into code (and still running into shape mismatches no matter how many networks I've built).

So I built BlockDL: a free and open-source tool to design neural networks visually in your browser. It updates live Python/Keras code as you build, and automatically checks for shape and connectivity errors in real time.

It supports: - Most common layer types (Conv2D, Dense, LSTM, etc.) - Multiple input/output networks - Skip connections - Real-time parameter and shape validation

In addition to the tool itself, BlockDL includes an interactive learning section with five free beginner-to-intermediate courses. These focus specifically on the architecture design process which is a gap I've noticed in most theory or code-heavy ML courses. Lessons are interactive and visual, with guided builds and optimization challenges.

The tool is fully free, and donations help fund my college tuition.

Would love to get some feedback.

Try it here: https://blockdl.com GitHub (core engine): https://github.com/aryagm/blockdl