0state builds the layer
other software trusts.
0state is me, Kam Low. Fifteen-plus years building protocol-level software and real-time systems, much of it still running in production. I take the AI work that is really an infrastructure problem and leave the chatbot wrappers to someone else. Not because the models are uninteresting, but because the foundation is where the work that lasts gets done.
What I build
AI Architecture & Orchestration
Designing how AI systems connect to business logic, data pipelines, and user-facing products. Not chatbot wrappers. The actual infrastructure that makes intelligent systems work under real conditions instead of in a demo.
MCP Development
Building Model Context Protocol servers and integrations. MCP is the interop layer models have been missing. mcp-parser snapshots and validates real servers; sourcey renders them straight into docs. The shipped work is the evidence.
Developer Tools & Infrastructure
APIs, documentation systems, build pipelines. The unglamorous work that makes everything else possible. My open source tools are still running in production a decade after I built them. That's not a flex, it's a minimum standard.
Product Engineering
For select clients building AI-native products. I don't take every project. I take the interesting ones.
How I work
One person, full ownership. No project managers between you and the person writing the code.
I start with architecture, not wireframes. The hard thinking happens before the first line of code.
I build things that last, not things that ship fast and die.
I say no to projects that don't interest me.
Background
I was building real-time communication protocols and cross-platform plugin systems years before the current AI wave. The infrastructure mindset is not new here; only the models changed.
Before the AI work I built icey (C++ media and networking), symple (a real-time messaging protocol with WebRTC signalling), sourcey (a documentation platform), and moxygen (Doxygen to Markdown). All still maintained. All still in production somewhere. The newest of it is scafld, a deterministic protocol for multi-phase agent work; it signs off on AI-written code with a receipt CI can re-check, and I run it on my own code before anyone else's.
The work is technical. The thinking behind it is not limited to technology. Good systems design means understanding how things actually work, the forces underneath the code as much as the code itself: biological, technological, political, psychological. That is what makes the infrastructure hold long after the trend that funded it is gone.
Get in touch
I take a few projects a year, and only the ones that are really infrastructure problems. If you are building something that has to hold under real load, send the architecture, not the brief, to [email protected].