BOQAI: Autonomous CAD-to-BOQ Engineering Copilot & Structural Takeoff Engine

In structural engineering consultancies and general contracting firms, quantity surveyors spend grueling days manually tracing beam spans, calculating gross versus net slab areas, tabulating column schedules, and deducting window openings. This workflow is labor-intensive and repeats every time an architectural revision is issued. When engineering teams attempt to pass raw CAD blueprints directly to generic Large Language Models, they encounter three critical roadblocks.

First, LLMs produce computational hallucinations that are fatal for high-stakes commercial tenders, because neural networks predict the next statistical token rather than executing strict numerical arithmetic. Second, feeding millions of raw coordinate floats into cloud AI APIs blows context windows within seconds and exhausts rate limits. Third, a 2D floor plan inherently lacks crucial 3D structural parameters such as story heights, slab thicknesses, concrete grades, and rebar density ratios, necessitating human engineering input before computation can proceed.

To eliminate AI hallucination entirely, I architected BOQAI on a modular hybrid pipeline that strictly isolates the AI model from mathematical arithmetic:

The platform first ingests binary AutoCAD files locally using LibreDWG, ezdxf, and Shapely. It converts spatial vector entities into a pre-aggregated semantic geometry summary, compressing 50MB CAD files into less than 3KB of structured metadata, achieving a 99.5% reduction in data volume.

Next, the AI copilot powered by Gemini 2.5 Flash or Claude 3.7 Sonnet inspects the structural drawing to detect missing engineering specifications required for estimation, such as foundation depth, blinding concrete thickness, clear story heights, and structural slab thickness. These requirements are presented through an interactive verification cockpit, allowing the project engineer to choose building code-compliant specifications with one click or enter custom project parameters.

Once parameters are verified, the workload transfers entirely to a pure Python deterministic math engine. The calculation engine computes concrete volumes, estimated rebar tonnages, masonry surface areas, and deductions for door and window openings with exact numerical repeatability and zero random approximations.

Beyond generating raw numbers, BOQAI includes an independent Layer-2 verification engine that benchmarks every computed line item against structural thresholds defined in the Saudi Building Code (SBC 301-304) and Egyptian Code of Practice (ECP 203):

The resulting takeoff delivers an auditable, interactive workspace tailored for technical directors, estimators, and site engineers:

The workspace supports four view layouts including balanced split view, full takeoff book, dedicated canvas focus with interactive zoom and pan, and comprehensive audit view, complete with real-time axis search and vector neon element highlighting upon row hover.

Upon finalizing the audit, the platform exports a professional four-sheet Excel workbook powered by native, live spreadsheet formulas rather than dead values. The exported file features an executive cover sheet, summary bill of quantities, detailed measurement takeoff ledger with grid axes and dimensions, and an official QA/QC audit certificate. Project stakeholders can adjust unit prices or dimensions directly in Excel, triggering automated recalculations across summary totals and tax schedules without breaking audit integrity.

In rigorous testing on residential villas and multi-story commercial blueprints, BOQAI demonstrated dramatic improvements over manual workflows: