Evis Drenova is a software engineer and former Neosync cofounder. He writes implementation-focused explanations of machine learning and privacy engineering, including small Rust projects that expose the machinery hidden behind training libraries.[1]
Start with Building an Autograd system in Rust →
X · evisdrenova 
GitHub · evisdrenova Career
Drenova says he cofounded Neosync and moved from full-time coding into sales, hiring and customer support as it grew. The platform anonymized production records for development environments. Its database-relationship handling kept references between tables consistent while replacing sensitive values and selecting subsets, so realistic test data did not depend on copying raw production information. His site reports the company’s acquisition by Grow Therapy.[3][1]
Contributions
His nanograd tutorial implements automatic differentiation in Rust. A forward computation records operations in a graph; a backward traversal applies the chain rule and accumulates derivatives for each parameter. Those derivatives guide weight updates. Building the graph and its gradient operations makes a training loop understandable beyond calling a library’s backward method. The project is educational rather than a complete high-performance framework.[2]
Tools, writing & talks
Writing, research & talks
Start here
A worked implementation of computation graphs, derivatives and a neural-network training loop.
Follow the scalar operations before the neural network. Basic Rust and the calculus chain rule help; the article explains how forward values become backward derivatives.
Author[2]
Tools
- Sophonsophon.at
Platform for discovering AI research papers, linking each to the benchmarks, tools, and models it introduces or uses.
Related projectOriginal source