About DatumOS

Construction data deserves a fixed point.

Every site has a datum — the mark all measurements refer to. We started DatumOS because the industry’s commercial data has no such mark: the same position is retyped, reformatted, and re-priced at every handover. We’re building the software that ends that.

The story

From a parser to a platform.

DatumOS began with a concrete problem: German construction tenders arrive as GAEB files that most modern software can’t read properly. So we built pyGAEB — an open-source Python engine that turns every GAEB version and exchange phase into one precise, typed model — and published it under MIT for anyone to use and audit.

Working with door and window manufacturers showed us where that foundation matters most: estimators spending days retyping tenders before they can quote. The DatumOS platform grew out of that work — upload a tender in any format, get a priced, reviewable, send-ready quote back in minutes.

The destination is bigger: one record every commercial workflow reads from — and specialized agents on top that prepare each of those workflows, from bid comparison to change orders, for human judgment. Germany first, because GAEB gives structured data a head start. Then everywhere estimates still live in PDFs.

Who’s behind it

LLM practitioners, close to the trade.

We build LLM applications for a living — the kind that have to be right on messy, high-stakes data, not the kind that only work in a demo. DatumOS turns that discipline on construction’s commercial data. The platform, the open-source engine, and every pilot are built by the four of us; the people you meet are the people who write the code.

Kalyanakannan Padivasu

Designs the LLM systems at the core of DatumOS — the pipelines that read a raw tender and return structured, priced, reviewable data. A decade building production Python platforms, and the one who decides where a model can be trusted and where a human stays in the loop.

Dhiyaneshwar Chandrasekaran

Owns the data foundation that makes the AI dependable: the ingestion, retrieval, and infrastructure that keep results consistent when tenders arrive in a dozen broken formats. Years of running high-throughput Python and distributed data systems in e-commerce and finance.

Swathi Narayanan

Makes the output hold up under scrutiny — the evaluation, edge cases, and quality discipline that let demanding customers rely on what the system produces. Eleven years keeping enterprise software correct where mistakes are expensive.

Sureshkumar Sreedharan

Turns model capability into workflows estimators actually use — the interfaces and product logic that put AI where the work already happens, without adding steps. Eleven years shipping production applications across the stack.

Talk to us directly.

No sales team, no ticket queue — you talk to the people who build the product. Whether you want a pilot, an integration, or just to sanity-check an idea about GAEB tooling.

Get in touch →