Systems, tools, and experiments

Try the software demos, or read about my video projects and the infrastructure behind them.

Switchyard

A workflow for assigning development and research to AI models, with separate reviewers and a record of the results.

Explore the workflow Launch terminal demo

OSINT feed

Live updates from public reports, grouped to reduce repetition. The feed keeps the last hour of posts and media.

How the feed works Open live feed

Atlas

Control devices, check the network, and research a question through a Signal conversation.

Read integration notes Launch Atlas demo

The Office: Blu-ray to 4K

I’m a huge fan of The Office. I built a pipeline to upscale my Superfan Blu-ray set to 4K, including its extra content. I compared models against the source footage and selected one based on how well it preserved the show’s grain and texture. The completed run covered all nine seasons.

The pipeline retained the original audio and subtitles and checked finished files before adding them to my library. I specified progress reporting so I could follow long runs without supervising every episode.

Analog video restoration

I developed a restoration pipeline for analog home video. It corrects color casts and reduces tape noise and ghosting before an AI model upscales the cleaned picture.

I kept those stages separate, with lossless intermediate files, so I could inspect what each step changed. A short sample has to pass review before a full recording is processed. Listening comparisons led me to a mild audio cleanup setting.

Infrastructure behind the projects

Outside work, I build and run Linux servers, virtual machines, containers, storage, and networks. The software is a mix of open-source and commercial tools that I connect to my services and devices.

AI agents now orchestrate and manage most of this infrastructure. I use deterministic scripts and tools to control when actions can proceed and to verify the results. I make the architectural and operational decisions, drawing on years of hands-on work before I used agents.

Recovery checks

I replaced a recurring manual network repair with a recovery check that follows the current gateway. We deliberately removed the route to test whether the check would restore it. A second run confirmed that it left the working route alone.

Hosting, monitoring and data

Application hosting: I host applications with Proxmox and Docker and manage storage, networking, resource allocation, and backups.
Service health: I use ntopng, Netdata, and NetAlertX with a dashboard that brings server, network, storage, and service health together.
Application data and service records: I use PostgreSQL for application data and SQLite for service history, caches, and searchable records.
Keyword and vector search: One of my projects collects live reports into SQLite, generates embeddings in background workers, and combines keyword and vector search. AI processing runs separately from collection, and search can fall back to keywords when the embedding service is unavailable.

For the career behind these projects, read about my work.