I build the tools I wish already existed.
I build practical tools around AI, self-hosting, media systems, automation, remote compute, native apps, and game modding.
Most of my projects begin with an annoying workflow, a missing feature, or a system that technically works but could work much better. I like turning those problems into tools that are fast, understandable, and genuinely useful outside of a demo.
My usual design goals are simple:
- Solve a real problem
- Keep the workflow practical
- Make important behavior visible
- Prefer user control over hidden magic
- Remove friction without creating new complexity
A lot of my projects start with the same thought:
“This should exist, so I’m making it.”
A native, Apple-first Jellyfin client for iOS
LibrarWUI is designed to make browsing and watching a self-hosted media library feel polished, responsive, and at home on Apple devices.
Instead of wrapping a web interface inside an app, LibrarWUI focuses on native navigation, playback behavior, interface design, and platform integration.
The goal is a Jellyfin client that feels intentionally built for iOS rather than merely adapted to it.
The app is currently in active development, with a public release and open-source release planned once the project is ready.
An AI-facing remote compute control plane for Vast.ai
VastAI-MCP is designed around a safe and inspectable workflow for finding, comparing, planning, and managing remote GPU compute.
The project aims to let AI agents work with Vast.ai without handing them unrestricted infrastructure access.
Current areas of focus include:
- Searching and comparing GPU offers
- Distinguishing secure cloud and community-hosted capacity
- Inspecting machine specifications and pricing
- Resolving templates and environments
- Attaching datasets
- Creating immutable rental plans
- Reviewing costs, risks, and warnings
- Approval-based execution workflows
- Spend limits and infrastructure safeguards
Experiments in local AI systems, coding agents, and post-training
HearthLM is an umbrella for my work around local language models and practical AI tooling.
The goal is not merely to run a local model. It is to build the systems around it that make it more capable, controllable, and useful for real work.
Areas of interest include:
- Coding agents
- Local model inference
- Tool use
- Context management
- Model evaluation
- Post-training and fine-tuning
- Dataset generation
- Long-running agent workflows
- Human approval and safety systems
I also build companion tools, workflow improvements, interface experiments, and modifications for games and custom systems.
These projects usually begin with some variation of:
“There has to be a better way to do this.”
Some current projects are still private or in active development.
AI-powered metadata enrichment for Jellyfin libraries
MetaMender scans Jellyfin libraries for missing or underdeveloped overviews, generates concise streaming-style descriptions with an AI provider, and writes the improved metadata directly back to Jellyfin.
- Supports movies, series, episodes, albums, artists, books, and other configured item types
- Detects missing or unusually short overviews
- Integrates directly with the Jellyfin API
- Supports hosted and locally running models
- Creates timestamped before-and-after logs
- Tracks input tokens, output tokens, and estimated cost
- Runs as a lightweight Python tool without requiring another server
- OpenAI
- Anthropic
- xAI / Grok
- OpenRouter
- LM Studio
- Ollama
- Generic OpenAI-compatible endpoints
Python · Jellyfin API · Hosted LLM APIs · Local LLMs
- AI tools that can safely operate real systems
- Native apps for self-hosted services
- Jellyfin and media-management tooling
- Local model workflows
- Remote GPU compute systems
- Automation scripts and workflow helpers
- Game and modding tools
- UI experiments and app concepts
- Small tools that eliminate disproportionately annoying problems
| Area | Technologies |
|---|---|
| Languages | Python, Swift, C#, C, C++, JavaScript, TypeScript |
| Web and UI | React, Next.js, Tailwind CSS, HTML, CSS |
| Systems and tooling | Node.js, PowerShell, FFmpeg, Supabase |
| Apple development | Swift, SwiftUI, iOS |
| AI development | MCP, coding agents, hosted LLMs, local LLMs, post-training, evaluation |
| Self-hosting | Jellyfin, media automation, local inference |
| Game development | Unity, MelonLoader, modding workflows |
Right now I am especially interested in:
- Building a better native Jellyfin experience on Apple platforms
- Creating safer AI-controlled infrastructure workflows
- Improving local models through post-training and evaluation
- Making remote GPU compute easier to search, understand, and control
- Building practical automation for self-hosted media
- Turning repetitive or frustrating workflows into focused tools
