Hello, I'm Silas Marvin, and I'm passionate enthusiast of Rust, Zig, and machine learning. Currently, I'm working at PostgresML, where we're focused on streamlining the development of machine learning applications within database systems.

Github | Twitter | LinkedIn | Newsletter


Current Projects


LSP-AI is an open-source language server that integrates AI-powered functionality into any editor with LSP support. It serves as a backend for performing completions with large language models and other AI-powered features.

Check out the LSP-AI GitHub repository for more details and to contribute.



Many of my talks involve live coding demos but unfortunately I only have slides to share here.

Maximizing RAG Performance with LlamaIndex on PostgresML

AI Dev Summit, San Francisco, 2024

An overview of the limitations and liabilities imposed by RAG and the solutions building on PostgresML provides.

Slides ->

Unified RAG

AI in Software Development Conference 2024

An introduction to RAG abstractions and building high performant RAG systems.

Slides ->

RAG Masterclass

Postgres Conference, San Jose, 2024

A crash course for SQL developers on building RAG applications from scratch.

Slides ->



Conway's Game of Life Through Time

Monday, April 1, 2024

Utilize Rust and the Bevy game engine to recreate a fascinating visualization of Conway's Game of Life through time.


Exploring Gleam with Genetic Algorithms

Friday, March 22, 2024

Meet Gleam, one of the most popular new functional programming languages by building a simple word-solving genetic algorithm.


Simple Automatic Differentiation in Haskell

Tuesday, May 16, 2023

Implement a simple automatic differentiation system in Haskell by storing tensor operations on a tape and using the State monad to efficiently calculate gradients through back-propagation.


DNNs from Scratch in Zig

Sunday, April 23, 2023

Explore building a simple deep neural network from scratch in Zig, utilizing only the standard library to create layer, activation, and loss modules that can be chained together to solve the MNIST dataset with 95.86% accuracy.


Chaos Networks Part 2

Monday, March 6, 2023

Building upon the previous post, this article delves into the refinements made to Chaos Networks, including enhancements to the automatic differentiation library, the introduction of batching for improved performance, and more.


Chaos Networks Part 1

Tuesday, January 17, 2023

Introducing Chaos Networks, a novel approach to feedforward neural networks that abandons the traditional layer-based paradigm in favor of a semi-randomly connected acyclic graph.


3d Cellular Automata

Sunday, November 6, 2022

Leveraging the power of Rust and the Bevy game engine, this post walks through the creation of a 3D cellular automata emulator capable of simulating up millions of cells.


Bevy dfdx and the Classic Cart Pole

Sunday, October 23, 2022

This post demonstrates how to solve the classic Cart Pole problem using Rust's Bevy game engine for visualization and the dfdx library for deep reinforcement learning.


© 2024 Silas Marvin. No tracking, no cookies, just plain HTML and CSS.