How does Bildstak answer questions in plain English?
Bildstak translates your plain-English question into queries across every connected project source, joins the results using project identifiers, and returns a coherent cited answer — no SQL, no exports needed. An AI analyst coordinates the query and structures the response.
When you type a question into Bildstak, a multi-step process happens in the background:
1. Intent understanding An AI analyst reads your question and identifies what data is needed — which systems to query, which identifiers to use as join keys, and what form the answer should take (a table, a number, a list of records, a narrative).
2. Query generation and execution Bildstak generates structured queries against each relevant connected source — the schedule, the cost ledger, the RFI system, the BIM model — and runs them simultaneously. The query language differs per source (SQL for federated data, REST API calls for Procore, IFC queries for the model) but that is invisible to you.
3. Federation and joining Results from each source are joined in the federation layer using the project's join keys — IFC GlobalID for BIM, activity IDs and cost codes for everything else. A result set that spans five systems is assembled as a single coherent dataset.
4. Answer synthesis and citation The AI analyst structures the joined results into a plain-English response. Every figure, every record mentioned in the answer is accompanied by a citation — the source system, the record ID, and where relevant the timestamp — so you can verify the answer directly.
The result is that you can ask 'what is the total cost of open RFIs on Package 2?' without knowing how Procore and SAP store their data, and receive a cited, current answer in seconds.
Updated 2026-06-19