An executable, compile-checked book teaching the structure under machine learning — tokens, embeddings, attention, and training modeled as typed transformations in Rust, so ML reads as explicit structure instead of framework magic. Ships runnable examples, exercises, and an open-access mdBook. Co-authored with F. Jafarranmani.

114 stars, 9 forks on GitHub; open-access book at hghalebi.github.io/category_theory_transformer_rs. Its underlying idea reached the Hacker News front page.

Active — public working draft

A governed Rust runtime for AI-driven software changes: explicit policy, bounded scope, verification receipts, and evidence-backed reports. Built so an agent's code changes are constrained, checked, and provable after the fact — the same "AI prepares, humans validate, audit trails prove" thesis applied to the software supply chain itself.

Open source on GitHub

Active development

A practical tutorial on wiring OpenTelemetry into Rig-based Rust agents — from first run to production-grade tracing rigor. Covers spans, metrics, and instrumenting LLM calls so agent behavior is observable and debuggable in production, not a black box.

Open guide on GitHub, 5 stars

Published

An mdBook and Rust companion project on building reliable long-running AI agents with Rust, PostgreSQL, and Rig — durable state, recovery, and orchestration patterns for agents that must run for hours or days without falling over.

mdBook and Rust companion for reliable long-running AI agents with Rust, PostgreSQL, and Rig.

Active

An open-source course teaching machine learning from first principles in Rust — treating ML as a translation problem from plain English to algebra to Rust newtypes to composable maps. Spans 11 modules from a single neuron to transformers, language modeling, systems, kernels, and inference, with tested companion crates for each stage and a CS336-equivalent Rust track. Live companion site at rust-ml.com.

10 stars, 51 commits; 11 lesson modules each backed by a tested, runnable Rust crate. Live at rust-ml.com.

Active, open source