Documents and drawings
- Construction document and drawing review
- Specification compliance review
- Drawing revision coordination
- Tender obligation and addendum review
Agentic AI × computer vision × construction
AI agents and computer vision for construction operations.
State Method designs and builds custom AI agents and computer-vision systems for construction companies, contractors, consultants and ConTech product teams. The work starts with one document-heavy or visual operational workflow and takes it from evidence to a controlled pilot and, when justified, a production system.
These are not unrestricted autonomous employees. Each system operates inside defined permissions, deterministic checks, named human approval, traceable events and a recoverable failure path.
- Market
- Construction and ConTech
- Starting point
- One bounded workflow
- Entry engagement
- 10-day Diagnostic
- Build path
- Pilot before production
Construction workflows
Build around the work your teams already do.
The best starting point is a repetitive process with representative documents or project evidence, a measurable manual baseline and a named person who owns the outcome. State Method maps the complete path rather than adding a model to one isolated step.
Coordination and intake
- RFI and submittal intake
- Email and attachment classification
- Project-record validation and routing
- Meeting actions and design-comment coordination
Site evidence and reporting
- Site progress reporting from notes, photos or video
- Inspection and test-record preparation
- Defect, NCR and safety-observation workflows
- Evidence-linked report drafting for human review
Commercial and handover
- Estimating and quantity-review assistance
- Supplier and quotation comparison
- Variation and payment evidence preparation
- Handover document completeness checks
The workflow does not need to match an existing example. Assessment begins with its real inputs, systems, exceptions and approval boundary.
Flagship demonstrations
Three agents show the complete construction AI story.
These interactive reference workflows use synthetic project evidence. They demonstrate system behavior and control patterns rather than client deployments or guaranteed performance.
Drawing Review Agent
Reads a controlled drawing package and schedules, surfaces revision conflicts with source references, runs fixed checks and waits for an authorized review decision before issuing a record.
Output An evidence-linked drawing review with visible exceptions and approval.
Explore the drawing agent →Site Vision Agent
Turns site notes, photos and video into a structured progress-report proposal while preserving location uncertainty and Site Manager authority.
Output A reviewable site report draft linked to the available visual evidence.
Explore the site vision agent →Construction Workflow Agent
Checks incoming RFI and submittal emails, attachments and project references, prevents duplicate or invalid records and routes exceptions without answering or approving them autonomously.
Output A checked and routed project record with a reconstructable decision path.
Explore the workflow agent →Agent definition
The model interprets. The system controls.
For State Method, an AI agent is a bounded software participant. It can interpret approved project information, prepare a proposal and use explicitly permitted tools. It cannot silently expand its own authority or treat a plausible answer as an approved action.
Model interpretation
Extract, classify, compare, summarize, identify uncertainty and prepare a cited draft or recommendation.
Output A bounded proposal linked to its available evidence.
Software authority
Validate inputs, enforce business rules and permissions, control state transitions, record events and block invalid actions.
Output An observable workflow that only advances through valid states.
Human responsibility
Review consequential outputs, resolve exceptions and approve, correct, reject or stop the proposed action.
Output A named owner remains accountable for the operating decision.
Recovery
Use timeouts, safe retries, idempotency, escalation and manual fallback to return failed work to a known state.
Output Failure is visible and recoverable instead of silently passing as complete.
Reference architecture
One controlled path from project evidence to action.
The exact architecture depends on the workflow and existing systems. Retrieval, a knowledge graph, MCP or another agent protocol is used only when it solves a demonstrated requirement—not as architecture theatre.
- Receive authorized drawings, specifications, photos, video, messages or project data.
- Use document parsing, computer vision or multimodal models to structure relevant evidence.
- Retrieve bounded project knowledge and preserve source references.
- Let the agent prepare a proposal, finding, route or draft within its assigned task.
- Run deterministic validation, business rules, permission and tool-access checks.
- Route consequential or uncertain cases to the named human approval owner.
- Execute only the approved action through an authorized integration.
- Record evaluation results, state transitions, decisions, failure and recovery events.
A workflow may stop earlier in this path. Not every use case needs retrieval, tools or external action.
System scope
Connect project evidence to a reviewable outcome.
Inputs
- Drawings, specifications and schedules
- Emails, forms and attachments
- Site notes, photos and video
- Registers, rates and structured project data
Integrations
- Approved document and data sources
- Existing internal tools and APIs
- Sandboxes or staging environments for pilots
- Authorized output and notification channels
Controls
- Input and schema validation
- Role and tool permissions
- Deterministic business rules
- Approval, audit events and safe stops
Outputs
- Evidence-linked findings and drafts
- Checked and routed project records
- Exception queues for named reviewers
- Approved artifacts with decision history
Verified delivery experience
Published tools and integrations, not a logo wall.
The public project evidence documents production integration across construction plan processing, AI pipelines, application software and AWS infrastructure. State Method names a platform only when a linked source supports the claim.
Construction plan processing
- Kreo wall, door and polygon extraction
- Shapely geometric processing
- PMR accessibility and fire-safety rule checks
- Floorplan polygon visualization
AI and computer vision
- OCR and large-document processing
- YOLO detection and room-mask pipelines
- Vision-language model classification
- Gemini construction-site video analysis
Application engineering
- Python and FastAPI services
- PostgreSQL data flows
- Project-context routing to AI-agent skills
- Review interfaces and structured report generation
Production infrastructure
- AWS Lambda and SageMaker
- AWS ECS, ECR and load balancing
- Production API deployment
- Model training and inference infrastructure
No production experience with an unlisted AEC platform is implied. A platform is named only when a linked project source verifies the integration.
Enterprise readiness
Data, access and ownership belong in the system design.
Enterprise readiness is defined against the client’s actual environment rather than implied by an AI demo. The Diagnostic records the requirements. The pilot tests the smallest credible operating boundary before production hardening.
Data boundary
- Authorized and representative inputs
- Data classification and retention requirements
- Rules for model-provider and third-party access
- No confidential documents through the public assessment form
Access and security
- Least-privilege roles and tool permissions
- Separated development, pilot and production environments
- Explicit approval for consequential actions
- Observable failures and incident routes
Hosting and integration
- Architecture chosen around existing client constraints
- API capability and access reviewed before commitment
- Sandbox or staging connections used during a pilot
- Production reliability scoped only after evidence
IP and operations
- Code, data and deliverable terms set in the project scope
- Named owners for review and production operation
- Logging, evaluation and change-control requirements
- Recovery, runbooks and handover defined for proven scope
This page does not claim security certification, formal software partnerships or prior integration with a named construction platform unless a linked case study documents it.
Relevant evidence
Construction experience, with its limits kept visible.
State Method is led by Hugo Sequier. His documented construction work covers technical-document intelligence, floorplan computer vision, deterministic verification, site-video interpretation, report drafting and production software. The selected-work pages preserve the original project context and source-reported results.
AnalyzTech
A document-intelligence and floorplan-analysis platform connecting model interpretation, verification actions and a review interface. The source reports 80% time saved on verification workflows in that historical project context.
Floorplan automation
A production pipeline connecting PDF intake, polygon extraction, deterministic accessibility and fire-safety checks, and reporting. The source reports a 30% productivity increase in that historical project context.
Video to Report
A multimodal pipeline that extracts evidence from construction-site video and prepares a structured report draft for office review. The source reports about three hours saved per report project.
Evidence boundary
These projects were led by Hugo before State Method launched. Their results are evidence of relevant engineering work, not forecasts or guaranteed outcomes for another company.
Engagement path
Move from a use case to production without skipping the evidence.
01 — Diagnostic
In 10 business days, bound one workflow, inspect representative evidence, define permissions and recovery, and decide whether a pilot is justified.
Output Five decision artifacts and a reasoned go/no-go recommendation.
02 — Controlled Pilot
Build the smallest useful version with real tools in a controlled environment. Test model behavior, software checks, human review and failure recovery against acceptance criteria.
Output Working evidence for a production, change or stop decision.
03 — Production System
Harden only the proven scope. Define production access, integrations, monitoring, evaluation cadence, incident handling, recovery and operational ownership.
Output A bounded production system with explicit operating responsibility.
Progression is not automatic. A useful result can be to narrow the design, gather better evidence, keep the process manual or stop.
FAQ
Before the next decision.
Does State Method build custom AI agents?
Yes. State Method can design and build a custom agent around one defined construction or ConTech workflow. The scope includes the surrounding software controls, integrations, review interface, logging and recovery behavior required to operate it responsibly.
Do you work with contractors as well as ConTech companies?
Yes. The relevant buyer may be a contractor, consultant, technical services company or ConTech product team. The important requirement is a bounded operational workflow with an owner, representative evidence and a result that can be reviewed.
Can an agent connect to our existing construction software?
Potentially. State Method first reviews the available APIs, permissions, data quality and action risks. A pilot uses only the smallest set of integrations needed to test the workflow safely.
Will the agent replace professional review or sign-off?
No. State Method does not offer AI professional or regulatory sign-off. Models may prepare findings or drafts, while the appropriate named person retains authority for consequential review and approval.
Do we need an existing AI prototype?
No. You need a real workflow, an operating owner and representative examples. An existing prototype can be reviewed, but the decision depends on its evidence rather than how mature it is described.
Continue with the evidence
Review the next relevant boundary.
Construction AI assessment
Bring one construction workflow.
Describe the current process, the documents or project evidence involved, the expected outcome and the person who must review it. Do not send confidential material through the public form.