NeuralNet
Design neural networks visually and generate Keras code
NeuralNet is a visual model-building prototype. Users connect layers in a React canvas and turn the graph into inspectable Keras code. The current hosted training view uses simulated metrics.
A visual graph must reject invalid connections and produce code in a consistent order without relying on unsafe runtime evaluation.
The editor stores layers and connections as graph data. The backend checks their order and generates Python source through an AST-based compiler. The current deployed training screen demonstrates the intended workflow with simulated metrics.
The live prototype supports visual architecture design and Keras code generation. Training and deployment are represented in the interface, but the hosted training route currently runs in mock mode.
Technical Specs & Highlights
Built a React Flow editor for visual layer composition.
Implemented topological sorting and AST-based Keras code generation.
Created dataset, training, and deployment interface flows.
Made the current demo state clear: hosted training metrics are simulated.
Looking for this skill set?
Jayant is open to full-stack engineering roles, internships, and high-impact software challenges.
AstroPixel
AstroPixel turns large astronomical images into browser-friendly tiles. I built the upload pipeline, deep-zoom viewer, annotations, and access controls from end to end.
Krypton
Krypton brings arXiv and OpenAlex results into one search flow, ranks papers with visible criteria, and uses Gemini to help readers summarize and explore the literature.