Staying in sync
Genlock
Timing, transport, networking. If a feed falls apart mid-stream, Genlock wants to know why, and whether it had to.
Applied R&D for AVoIP and AI, encoding, and provenance pipelines.
CueSeal researches how video gets made, moved, saved, and archived, then publishes what it finds. Including the parts that didn't work, or barely scraped by.
01
Who we work for
A lot of broadcast and video engineering got worked out decades ago, inside facilities with real budgets and full crews. The current stack is drifting between four eras of technology: analog, digital, IP, and now the newest frontier, AI. Cameras got cheap and computing got fast, but most of that old discipline never really made the jump to the people now doing the work with neither.
CueSeal pokes at those gaps out of sheer purpose-driven curiosity after reflecting on years on the facility side, inside smart homes, home studios, and professional media houses. These days, that means spending weekends testing how much of it still holds up when the budget is a credit card and the control room is a spare bedroom, with agents doing more of the actual test-running than they used to. We want to make it work for everyone doing media work with limited resources.
02
Three tracks
What quality actually costs
MuxNode
Encoding, packaging, delivery. Most encoding advice online is a preset somebody got lucky with once. MuxNode pokes at it until something actually checks out.
What survives the trip
Cuebus
Metadata, provenance, preservation. Captions, color data, credentials, camera metadata: the stuff that quietly falls off a file somewhere between capture and upload, and nobody notices until it matters.
03
An agentic workflow processes research and data
A handful of AI agents are built and maintained for this effort to run research sweeps, put together weekly digests, and check whether a link still works or a spec quietly got a new revision. That workflow buys back the hours an engineer would otherwise spend on upkeep. We make sure AI works hard for us, not the other way around.
Every actual measurement, every line of code, every conclusion gets made by a person and signed off before it goes anywhere. An agent can draft a paragraph. It can't tell you a number is true. Somebody has to check, then put their name on it, and own the findings.
There's a second experiment running underneath the first: whether a small team (currently one engineer) using agents to soak up the busywork alongside a day job, can produce research independently. This site, and its projects, exist to find out whether decentralized research and development can be done independently in the era of AI.
04
Open by default
If it can be published, it gets published: the method, the code, the raw numbers, the fixtures, and the experiments that went nowhere. Even when we mess up, we trust you'll let us know. That is, after all, the fun of open source research and development.
Standards get cited, never copy-pasted somewhere they don't belong. If something here leans on a proprietary tool, that gets stated clearly. A result you can't check is worth exactly what you paid for it.
05
Recent work
Where this actually stands
Nothing's published yet. Term one runs through December 2026. Work in progress goes up on the research page as it happens, which moves faster than this page does.
06
Say hello
No newsletter, because nobody needs another one. A Discord is under construction for anyone curious about our work. We welcome any and all media technologists, or technologists in general.
Everything else goes straight to Mark at mark@cueseal.com.