GradieX

GradieX is a lightweight neural network engine written in modern C++ with a companion web front-end.

Current Core Capabilities

• Dense feedforward neural network with configurable hidden/output activations
• End-to-end forward pass across multiple hidden layers
• Backward pass with gradient propagation and per-parameter updates
• Loss functions: mean squared error and binary cross entropy
• Training loop with epoch logging and inference API
• Built-in XOR training example in main.cpp

Build and Run on Linux

Compile:

g++ -std=c++17 -O2 -Wall -Wextra -pedantic main.cpp -o gradiex-linux-x86_64

Run:

./gradiex-linux-x86_64

Distribute Binary via the Web App

To make the Linux binary downloadable from the frontend app:

mkdir -p frontend/public/downloads
cp gradiex-linux-x86_64 frontend/public/downloads/gradiex-linux-x86_64

Then start/build the frontend:

cd frontend
npm install
npm run dev

The binary is served at:

• /downloads/gradiex-linux-x86_64

Project Layout

NebulaNeural/
  main.cpp
  README.md
  frontend/
    public/
    src/

Next Steps

• Add mini-batch training and shuffled epochs
• Add model save/load for trained weights
• Add additional examples (regression and multi-class classification)