Construction PM’s AI assistant answers project questions in plain English or Spanish: costs, billings, profit, schedule, project health, and a “what’s on my plate” checklist of scope items. A bilingual training library answers documentation questions over your uploaded specs and SOPs, and an AI inspection on daily reports surfaces a Boss Report queue for the superintendent. Behind every answer is data the project manager and foremen have already entered.
From the chat panel, the project manager asks (or clicks a dropdown) for costs, billings, or profit. The assistant pulls from labor logs, cost logs, draw requests, and budget items already in the system, and returns the answer with the underlying numbers. The cost view shows budget against spent with weekly burn rate and a 7- and 30-day velocity. The billing view shows progress billing against the schedule of values, retainage held, and the next AIA-style application’s draw amount. The profit view sums paid draw requests and subtracts logged labor and material costs into per-job and monthly views.
The schedule dropdown surfaces upcoming mobilizations, crew assignments, weekly forecasts, and conflicts. The agent answers questions like “what does next week look like?” and “who is mobilizing where?” with a structured response built from the mobilization periods, crew assignments, and labor history project managers and foremen have already entered. Forecast hours are simple averages from recent labor history, not a model prediction.
The project health dropdown is a portfolio view of risk: jobs land in on-track, at-risk, or critical based on cost and schedule performance, with a recommended weekly velocity to recover the at-risk ones. The “what’s on my plate” checklist (typed into chat, not a dropdown) returns scope-level checklist items still unchecked, grouped by job and sorted by next mobilization date. Both views read from data already in the system: budgets, schedules, labor and cost logs, and the scope checklists the project manager maintains.
The training library is a per-tenant collection of specs, SOPs, manuals, and training docs that the AI can answer questions against. Project managers upload PDFs, Word and PowerPoint files, spreadsheets, and images, plus web URLs and YouTube links (transcripts fetched automatically), tagged by CSI division and marked either “Human Use” or “AI Reference Standard.” A question typed in English or Spanish runs hybrid search (keyword and vector) over the AI Reference Standard corpus, and the answer is synthesized in the same language the question was asked, even when the source documents are in English.
Daily report photos and PDFs uploaded by foremen run through an AI inspection that scores each report and flags specific findings: wet adhesive on a walking surface, housekeeping and trip hazards, and similar safety or quality issues. The Boss Report is the queue of unreviewed daily reports waiting for superintendent review; the superintendent opens each report, reads the AI verdict and confidence score, and clicks Agree, Disagree, or Other on each finding. The decisions are recorded against the report so the review is auditable.
The AI assistant answers project questions in English or Spanish across costs, billings, schedule, profit, project health, and a checklist of unchecked scope items. It also inspects daily report photos and surfaces a Boss Report queue of unreviewed reports for the superintendent, and answers questions over an uploaded training library of specs and SOPs. Every answer is built from data the project manager and foremen have already entered.
No. The project manager enters and reviews the data; the AI summarizes, projects, and flags. For example, foremen upload daily reports and photos, the AI flags safety and quality findings with a confidence score, and the superintendent agrees, disagrees, or marks Other on each finding. The decision stays with a person.
Two ways. First, the AI answers from data already in the system (labor logs, draw requests, mobilization periods, scope checklists, and so on) rather than from open-ended generation, and the answer cites the underlying numbers. Second, the training library is the only source the AI references for documentation questions, and the AI Reference Standard tag controls which documents the search includes.
