Signal Engineering
8th layer protocols.
Systems work for people who want to understand what they have built: local-first software, a second look at architecture, and advice on training neural networks and on working well.
What I do
Local-first systems. Desktop applications and infrastructure that run without a cloud account: encrypted local storage, sync you can audit, data that stays in the building. Current work: a GDPR enforcement tool and a document pipeline that reads, classifies and files years of paper.
Analysis and architecture. The slow part. I sit with an existing system until its actual structure shows, then draw the version that holds: integration boundaries, storage decisions, and a record of why.
Models as components. Language and vision models where they earn their place: classification, extraction, a first pass that a person confirms, and a verification gate wherever an error would be costly. No chat window for its own sake.
Training neural networks. Advice from the data side: what the labels actually say, splits that do not leak, a baseline before an architecture, an evaluation that would survive a skeptical reviewer. Sometimes the useful answer is that nothing needs training.
Mental health and sustainability. I advise companies on both. The question is the same one as in the systems work: what an organisation asks of the people inside it, and what it can keep asking for the next ten years.
Working practice. Coaching for people who build alone or in small teams: deciding what to do next, finishing what is half-built, keeping a project honest about its state.
Articulation. Early systems often work before they can say what they do. I help research prototypes and products state their actual function, first technically, then publicly.
What runs under all of it is attention: what a system asks of the people inside it.
Context and clients
My work has moved through industry, research and independent practice: platform engineering and security at Keelvar, data science on the NEAR Protocol at Pagoda, enterprise systems through kuehlhaus for clients including Roche and Freudenberg. Alongside that I work with academic researchers, artists and independent practitioners.
About
I am Daniel Herrmann. I started in data science and moved through blockchain development into software that stays on the user's own machine. In 2018 a hackathon team of mine won first prize for business model at Sensors & {data} in Heidelberg, with a medical-data idea I have since argued against, in public, at length.
Next to client work I make generative art, write about psychology and attention, and build tools: Schwellenkarten, a card deck for self-inquiry; menon, two instances of one language model questioning each other; Sicherlich, a vault for your own medical records. The art and the engineering are not separate practices for me. They ask the same question from two sides.
Working together
I take on a small number of engagements per year. Most begin with a structural examination: fixed scope, two to four weeks, written findings, and an architecture you can act on with or without me. Some engagements continue into the build. Some end there, on purpose.
To begin, write to hello@scopal-affairs.com: two sentences on the system, one on what is stuck.