Compare site evidence with the construction programme

Start with the operational bottleneck

A progress meeting can contain several different accounts of the same work. Subcontractor updates, photographs, programme entries, and site observations may refer to different locations or dates. The first improvement is to make those references consistent.

Choose one project area and a limited set of activities with observable completion criteria. Record what evidence is sufficient to mark an activity started, partially complete, complete, or unverified. A photograph alone does not establish concealed work, inspection acceptance, or payment entitlement.

Design the target workflow

Capture site imagery on a repeatable route and attach time and location metadata. Align the evidence with the project’s spatial model and activity identifiers. A computer-vision component can propose visible progress or discrepancies. A language model can turn reviewed findings into a concise coordination brief.

Give the project team the original capture, the model location, the proposed status, and the reason for a discrepancy. Accepted observations update the planning system through its normal approval path. Unclear captures become requests for additional evidence. Retain a history so the next meeting compares like with like.

Connect to the existing systems

Connection	Information	Control
Site capture	Dated imagery and locations	Use a consistent route and capture standard
BIM and programme	Model elements and activity IDs	Keep revision alignment explicit
Planning workflow	Reviewed observations and actions	The project team accepts schedule changes

Use stable identifiers and versioned inputs. A retry should resume the same piece of work, with the original permissions and evidence attached. Treat an upstream outage as a visible workflow state.

Fit the model to the task

This is a combination of computer vision, structured comparison, and reporting. A text-only LLM does not measure installed quantities from photographs. Select the vision component on project-specific evidence; use a smaller model for drafting reviewed summaries rather than asking a frontier model to infer the entire site state.

Keep a clear human boundary

Site staff validate exceptions, occluded areas, and incomplete evidence. The project manager owns programme changes. Keep contractual acceptance and safety decisions within their existing professional processes. Never convert an unverified visual inference into an automatic certification.

Measure the complete operating cost

Include capture time, model alignment, review, licensing, storage, and planning integration. Compare repeated reporting effort and the usefulness of detected discrepancies. Avoid assigning all project-wide schedule changes to the reporting tool.

Pilot checklist

• Define observable completion criteria for selected activities.
• Align capture dates, model revisions, and programme identifiers.
• Measure false alarms and missed discrepancies.
• Record how accepted observations change decisions.
• Keep contractual acceptance separate from visual progress inference.

Agree on acceptance thresholds before testing, then compare the proposed workflow with the same work handled through the current process. Expand only after the operation can explain both successful and failed cases.

Further implementation detail

• Choose the lowest-cost model that meets the task requirements
• Serve a local model to a team
• Ollama’s local API reference