Zero-Dependency Machine Learning

A foundational stochastic gradient descent engine written entirely in Go. Created to demonstrate sub-millisecond mathematical operations without the heavy execution overhead typically associated with standard Python frameworks.

Why Go?

Standard Machine Learning frameworks (PyTorch, TensorFlow) are massive C++ monoliths wrapped in Python. While excellent for research, they introduce massive deployment complexities and cold-start Latency penalties in production microservices.

The Architecture

This engine implements the core mathematical primitives required for neural network training and inference directly in Go structs, without relying on cgo or external C bindings.

By keeping the execution entirely within the Go runtime, we achieve:

• Zero Cold-Start: Sub-millisecond initialization.
• Micro-Binary Deployment: The entire ML engine compiles down to a single, static binary under 10MB.
• Provable Math: No black-box library calls; every tensor operation is strictly typed and verifiable.