Use Case: BIM and Construction File Triage with Caliper

Construction projects generate a constant flow of 3D geometry data. BIM models export to OBJ and STL for fabrication. Site surveys produce LAS/LAZ point clouds. Coordination reviews share glTF models for lightweight viewing. Fabricators send back STL files for approval.

Every handoff is a place where things go wrong: wrong coordinate system, missing geometry, files too large for the receiving tool, corrupt data from a bad export. Finding these problems early saves days of rework.

Caliper gives AI agents the metadata to catch these issues at the point of exchange.

Common problems in BIM/construction file workflows

Export validation

BIM software (Revit, ArchiCAD, Tekla) exports geometry to exchange formats for downstream use. These exports frequently have issues:

  • Empty or near-empty exports — a filter was misconfigured and the OBJ contains 12 triangles instead of 12,000
  • Dimension anomalies — the model was exported in millimeters but the receiving tool expects meters, producing a building that’s 1,000x too large or small
  • Non-manifold geometry — boolean operations in BIM produce geometry artifacts that break fabrication tools

An agent with Caliper checks the export immediately:

curl -X POST https://caliper.fit/tools/caliper_auto_stats \
  -H "Content-Type: application/json" \
  -d '{"file_url": "https://project-files.example.com/structural/steel-frame-export.stl"}'

The metadata includes bounding box dimensions, triangle count, and manifold status. The agent compares against expected values:

  • Triangle count below threshold → likely a bad export filter
  • Bounding box 1000x expected size → unit mismatch
  • Not watertight → geometry artifacts from boolean operations

Point cloud triage

Site surveys and as-built scans produce LAS/LAZ point clouds that range from a few megabytes to multiple gigabytes. Before processing (registration, classification, clash detection), teams need to know:

  • How many points does this scan contain?
  • What classifications are present (ground, building, vegetation)?
  • What are the spatial extents (does this cover the expected area)?
  • Is the coordinate reference system set correctly?

Caliper extracts this metadata without downloading and loading the full point cloud:

curl -X POST https://caliper.fit/tools/caliper_auto_stats \
  -H "Content-Type: application/json" \
  -d '{"file_url": "https://surveys.example.com/site-scan-2026-03-28.laz"}'

An agent can triage a batch of point clouds, flagging scans that are missing expected classifications, cover an unexpected area, or have anomalous point counts.

Coordination model checks

Coordination workflows share lightweight 3D models (often glTF/GLB) between trades. Before a coordination meeting, an agent can validate:

  • All expected models are present — check that structural, mechanical, electrical, and architectural models were submitted
  • Scene structure matches expectations — correct number of nodes, meshes, and materials
  • File sizes are reasonable — a 2 GB glTF that should be 50 MB suggests an export problem

Fabrication file validation

When fabricators send STL files for CNC or 3D printing, the receiving team needs to confirm:

  • File is watertight (required for toolpath generation)
  • Dimensions match the design intent (bounding box check)
  • Poly count is within machine limits
  • No degenerate triangles (zero area faces)

Workflow: agent-driven file intake

File arrives (BIM export, scan upload, fabricator delivery)
    ↓
Agent calls caliper_detect_format (free)
    ↓ expected format?
Agent calls caliper_auto_stats
    ↓
Agent checks metadata against project rules:
  - dimensions within expected range?
  - geometry count reasonable?
  - manifold/watertight if required?
  - classifications present (point clouds)?
    ↓
PASS → file enters project document management
FAIL → agent creates issue with specific failure details

For daily scan uploads or batch fabrication submissions, caliper_batch_stats analyzes up to 10 files per request.

Why this matters for construction

Construction rework costs the US industry an estimated $65 billion annually. A significant portion stems from data quality issues that propagate through handoffs. Catching a unit mismatch in an STL export before it reaches a CNC machine costs nothing. Catching it after fabrication costs thousands.

Automated file triage doesn’t replace engineering judgment — it catches the mechanical errors (wrong format, wrong scale, empty export, non-manifold mesh) so engineers can focus on the design decisions that require human expertise.

Getting started

  1. Add https://caliper.fit/mcp/ to your agent’s tool configuration
  2. Define validation rules per file type and project phase
  3. Wire up your file intake workflow (project management system, cloud storage, email) to trigger the agent
  4. Use caliper_detect_format (free) as the first gate, then full analysis for files that pass

Guides: Getting Started with MCP | Getting Started with REST | What’s in an STL File