I build local-first AI tools, prompt-heavy workflows, and developer systems that are meant to be used, inspected, and improved.

Most of what I work on sits somewhere around DSPy, LM Studio, OpenAI-compatible local runtimes, RAG, prompt tooling, and repo-aware automation. Some of it is public, a lot of it is not, but the pattern is usually the same: make the workflow more useful, more inspectable, and less dependent on black-box magic.

• DSPy experiments and teaching tools
• LM Studio tooling and plugins
• prompt libraries, prompt workflows, and prompt testing
• local-first agent and developer environments
• repo mapping, structure-aware tooling, and workflow orchestration
• small utilities that make AI-assisted development less annoying

• DSPyTeach — turns source material into structured teaching briefs
• lms-llmsTxt — generates llms.txt-style artifacts with DSPy + LM Studio
• rag-v2 — an LM Studio RAG plugin
• dspy_workspace — experiments, utilities, and scratch space for DSPy work
• prompt-docs — reusable prompt assets and workflow patterns

There is also a larger private pile of work behind this profile covering:

• local-first AI development tooling
• prompt infrastructure and benchmarking
• code review and session analysis tools
• repo intelligence and planning tools
• dashboards, registries, and internal workflow systems

Most of it is in service of the same idea: AI tooling should be understandable, controllable, and useful in real workflows.

I usually prefer:

• local-first over SaaS-first
• tools you can inspect over tools you have to trust
• systems that help you move faster without hiding what they're doing
• practical workflows over polished hype demos

[![Tokscale Stats](https://tokscale.ai/api/embed/AcidicSoil/svg?sort=tokens&compact=1)](https://tokscale.ai/u/AcidicSoil)

• LM Studio: https://lmstudio.ai/dirty-data
• X: https://x.com/d1rt7d4t4