Open Source AI Systems
This page brings together my open source projects focused on artificial intelligence systems, language model infrastructure, and the software patterns that support reliable AI applications.
These projects are distinct from front-end web development libraries and general-purpose utilities. They focus instead on the systems layer behind modern AI workflows: retrieval pipelines, structured contracts, evaluation, observability, and other reusable building blocks that help developers move from simple experiments to more durable implementations.
The goal of this collection is not to chase novelty for its own sake. It is to build practical, understandable tools that make AI systems easier to design, test, and extend over time. Where possible, the emphasis is on modular architecture, portability, and straightforward interfaces that can support experimentation without locking projects into a single vendor or framework.
As this area grows, it will serve as a home for open source work related to retrieval-augmented generation, contract-based AI development, evaluation tooling, and other infrastructure-oriented projects. Some of these projects are intentionally minimal and starter-focused. Others will evolve into more capable systems over time. Together, they reflect an interest in AI not just as a set of models, but as an engineering discipline shaped by structure, reliability, and reuse.
Featured AI Systems Projects
RAG Starter
RAG Starter is a minimal Python boilerplate for retrieval-augmented generation. It is designed to give developers a clean starting point for building systems that combine language models with external knowledge.
Rather than tying the project to a single stack, RAG Starter uses an adapter-first design that separates core RAG logic from vector database implementations. This makes it easier to understand the architecture, swap components, and extend the system as requirements change.
The project includes the core elements needed to begin working with retrieval workflows, including chunking, embedding, retrieval, prompt building, and vector database adapters. It is intended as both a practical starting point and a reference architecture for developers exploring knowledge-driven AI applications.
Brandon Himpfen
Prompt Registry
Prompt Registry is a lightweight system for managing prompts as structured, versioned assets with built-in evaluation support. It provides a clear way to store, render, and test prompts using a consistent Python-based interface.
Rather than embedding prompts directly in application code, Prompt Registry introduces a registry layer that resolves prompts by name and version. This creates a separation between application logic and prompt logic, making systems easier to maintain and evolve over time.
The project treats prompts as first-class components of AI systems. Each prompt is defined in a structured format, supports variable validation, and can be paired with evaluation cases to test behavior across changes. This makes it possible to move from informal prompt iteration toward a more repeatable and measurable workflow.
Prompt Registry fits naturally alongside other infrastructure-oriented tools such as contract layers, evaluation pipelines, and observability systems. It is especially useful in projects where prompt stability, traceability, and iteration discipline matter.
The design is intentionally minimal. It avoids heavy abstractions in favor of clarity, portability, and ease of integration into existing systems. It can be used as a standalone tool or as part of a broader AI systems stack.
Brandon Himpfen
How This Collection Relates to the Broader Software Page
This page is a focused extension of the broader Open Source Software collection. While the software page includes larger-scope projects across different technical domains, this section narrows in on artificial intelligence systems and the infrastructure patterns behind them.
Brandon Himpfen