AEC AI Lab
The bench where FastBid24 products come from. Every card is real work — including where it still fails. We publish the failures because a lab that never fails isn't running experiments; it's running marketing.
Door schedules are often rendered images inside PDF sheets — no text layer, merged cells, revision clouds. The extraction pipeline reads the rendered schedule visually and as parsed text, then reconciles the two before structuring rows.
Hardware sets live in the spec; openings live in the schedule. The mapping layer joins them opening by opening — and refuses to guess when the documents are silent, flagging instead.
Every extracted entity is checked against every other source that mentions it: schedule vs. spec, spec vs. plan, set vs. opening. Disagreements become review items with the evidence attached.
Every flag carries its evidence: what the schedule says, what the spec says, what's missing. The pipeline drafts the RFI with references attached, ready to send.
The extract → map → reconcile → verify pipeline is trade-agnostic by design. Seven trade products are in development on the same foundation, prioritized by waitlist demand.
The reconciliation layer already finds disagreements between documents. The next step is finding disagreements between scopes — requirements that appear in the project but land in no one's bid.
The loop
Schedule extraction, hardware mapping, reconciliation, and RFI drafting all started as “can AI even do this?” — and shipped inside Doors & Hardware AI. The multi-trade pipeline is in the prototype stage now; waitlist demand decides what graduates next.
The next experiment could start with the workflow your team hates most.