KoldOps · Industrial AI Research
Context infrastructure
for physical industry.
An industrial AI research lab building the substrate that connects language models to real operations.
IncWorxThe Thesis
The AI wave stops at the factory door until someone builds the substrate for it.
Every industrial operation runs on artifacts a language model cannot read. Engineering drawings in PDFs. Field notes on paper. Permits in scanned binders. Daily reports typed into email. Between the model and the operation sits a missing layer, and that layer is where the value is.
The winners in industrial AI will not be the labs training bigger frontier models. They will be the labs building the retrieval, the context engines, the protocol servers, and the agentic loops that make a frontier model useful on a real jobsite.
That is the layer we build, under contract with industrial partners who fund the research. The outputs deploy in the field and publish as public substrate.
Research
Four active directions.
Each direction runs against a specific industrial substrate. Progress compounds across them; every partner engagement touches at least two.
Context infrastructure
Storage, retrieval, and assembly for industrial artifacts. Beyond RAG over PDFs: drawings, permit trees, DDRs, spec hierarchies, jobsite conditions, portfolios of real property.
Protocol servers for physical operations
Model Context Protocol servers that expose ERPs, MES, CMMS, and field-data systems to agentic models, with the safety envelope those systems require.
Agentic operations
Long-running agents that plan, retry, and hand off across the analog handoffs of a real operation. Not chat. Not demo. Production loops that survive a plant shift.
Substrate audit methodology
How to measure whether a shop, site, or portfolio is ready for the model layer. Public methodology, applied under contract, sharpened by every deployment.
Field Work
Deployments running in production.
Every research thread lands in the field before we publish about it. Current partners run systems the lab built on their own infrastructure.
Field-data substrate replacing 1,000 hours of weekly manual handoffs across the quote-to-cash chain.
Real-time shop-floor context for a build-to-order operation. Station scans replaced paper travelers.
Unified operational context across three siloed systems. Same data everywhere.
Engagements
Two ways to work with the lab.
We take on a small number of new partners each year. Every engagement produces both a deployed system and a piece of the public substrate.
Research Partnership
A small number of industrial firms fund a research program with the lab. Every quarter ships a deployed system and a piece of published substrate. The right fit if you want the lab embedded in your operation and the outputs to compound over time.
Research Contract
A scoped engagement against a specific research question tied to your operation. The right fit for a bounded problem the operational-tooling market cannot solve, or as an entry point before a longer partnership.
Tell us about the operation and the problem inside it.
We take on a small number of new partners each year. Write to the lab with the shape of the operation, the research problem you see inside it, and how you would know we were making progress.