Construction ops, wired into one chat.
Pre-Series-A construction fit-out · ongoing
The problem
Construction projects do not lose time on the worksite — they lose it in the gaps between drawing, meeting, and status update. The workflow is mostly offline, so drawings, site updates, and meeting notes never connected.
What we built
CAD and PDF drawings, queryable
Drawing intelligence
Drawings parsed end-to-end. The system asks clarifying questions where a drawing is ambiguous, then feeds answers to planners and supervisors. Drawings become something the team interrogates, not inert documents.
One video replaces the unfilled form
Daily site video → status
Each supervisor records a single daily site video. Per-area status extracts automatically; the daily progress report writes itself, asking the supervisor only where it cannot tell.
Self-updating project tracker
Meeting summary → tracker
Every check-in recorded in context. Action items extracted and routed; the project tracker updates from the conversation itself — no minutes to type.
One interface over all of it
One chat assistant
Supervisors and ops query everything in natural language — site progress, drawings, schedules, action items — through one assistant sitting on top of the drawing, video, and meeting layers.
Outcomes
The transferable lesson
The hardest place to deploy AI is where the workflow is mostly offline. Start with the artifacts the team already produces — a drawing, a video, a meeting — and route them through the channel they already use. The interface decides whether the system gets used at all.