Towards Data Science is a community-powered publication that showcases work in data science, machine learning and artificial intelligence. Every day newcomers, seasoned researchers and industry practitioners publish tutorials, research notes and real-world case studies that help the field move forward.

Towards Data Science

An experiment using local LLMs (Mixtral 8x7B) running on a MacBook Pro M3 to automatically generate and optimize matrix multiplication algorithms in Rust. The author created a multi-agent system using Microsoft Autogen with five specialized agents (Proposer, Verifier, Coder, Tester, Manager) that iteratively generate, verify, and benchmark Rust code. Starting from a Strassen-like algorithm, the system evolved to produce NEON SIMD-optimized code with Rayon parallelism, achieving a 50% speedup over standard implementations. The workflow uses a vector database of 50 academic papers on matrix multiplication, semantic chunking for context management, and demonstrates that consumer-grade hardware can discover meaningful algorithmic optimizations through agentic AI workflows.

Using Local LLMs to Discover High-Performance Algorithms

Prepare the input data and vector database

The core code: autogen-core and GGML models