Fast screening
Scores · reasons · cited pages
Pursue · review · rejectHow GAIA Civil works
GAIA Civil separates a 3-minute fast-screening lane from comprehensive analysis. ORION moves selected-or directly submitted-project sets through a five-phase workflow for scoping, multimodal analysis, verification, and source-backed reporting.
Workflow map
The workflow branches at Project Hub. Screening is an optional low-cost lane; comprehensive analysis can begin directly or after a document is promoted. Both paths converge on a source-linked report and explicit human approval.
Scores · reasons · cited pages
Pursue · review · rejectMultimodal ingestion · typed stages
Needs · evidence · measurementsFunctional architecture
Approved project sources and customer intelligence produce page artifacts, retrieved evidence, typed needs, measurements, calculations, and a governed report. The technical reference explains each stage and its failure behavior.
Tenant-customized intelligence
ORION is not a generic document chatbot. The GAIA Civil tenant environment provides the proprietary products, engineering boundaries, discovery strategy, estimating rigor, permissions, and handoff rules used by the workflow.
Proprietary SKUs, approved products, application guidance, technical data sheets, competitor matrices, and approved-equal context define the portfolio ORION evaluates.
Structural, hydraulic, soil, material, scale, waste, overlap, conversion, QA, and discipline-specific skill modules define how technical evidence is interpreted.
Direct specifications, competitors, alternate language, commercial signals, design conditions, exclusions, and scoring thresholds define what deserves attention.
BALLPARK, TAKEOFF, or QUOTE_READY rigor, named reviewers, role permissions, report expectations, approval gates, and integration destinations connect output to accountable people.
Customization increases relevance; review states preserve authority. ORION applies customer-approved context and accuracy profiles, while GAIA Civil keeps source, confidence, assumptions, and required approvals visible.
Five connected phases
Each phase has a system responsibility, a retained artifact, and a human responsibility. Open the detail within every phase to see the handoff.
Select project documents, product catalogs, engineering references, application rules, competitors, qualification thresholds, estimating profile, and responsible reviewers.
Construction documents use varied terminology and distribute requirements across drawings, specifications, schedules, and addenda. ORION begins by defining which record is authoritative and which customer context it may apply.
Run a 3-minute opportunity screen across 500+ page project sets to calculate Technical Fit, Pursuit Readiness, Evidence Confidence, and a pursue/review/reject recommendation.
The screening lane checks configured product and application terms, incumbents, commercial/design/quantity signals, exclusions, text coverage, and optional semantic page matches without running the full generative analysis path.
Process text and page visuals, retrieve project/product/engineering evidence per need, assess scale and measurements, and run typed specialist analyses before BOM reconciliation.
Full ingestion creates page-level text, image, PDF, visual-summary, component, and provenance artifacts. Project Analysis V2 then resolves a stable evidence snapshot and fans targeted work through specialist roles.
Open cited drawings and specifications beside each finding, review highlighted regions, and inspect scale, formula, assumptions, cross-checks, methodology, and verification status.
The report distinguishes pursuit confidence from product and quantity confidence. Missing dimensions, contradictory sources, weak matches, and unresolved decisions remain visible rather than being smoothed into one answer.
Deliver the Qualified Opportunity Report, engineer verification packet, structured BOM outputs, technical inquiries, and approved API, webhook, MCP, JSON, CSV, or PDF handoffs.
The core report exists before optional actions and drill-downs are generated. Recommendations remain selective, can be empty, and route back into review rather than silently executing an external action.
Inside full analysis
The application reports these stages as analysis runs. Each stage produces a different class of inspectable evidence for the report and the reviewer.
| Stage | System responsibility | Record created |
|---|---|---|
| Scope | Resolve project and document scope; build stable citation context. | Project snapshot, source manifest, coverage, enabled skills |
| Scale & measurement | Assess documented quantities, scale evidence, geometry, and input gaps. | Measurement audits, calculations, assumptions, confidence, questions |
| Evidence collection | Retrieve project, product, and engineering evidence for each need. | Need-linked evidence groups and source citations |
| Engineering analysis | Apply engineering rules, calculations, and enabled domain methods. | Typed engineering analysis and calculated quantities |
| BOM reconciliation | Match product recommendations to authorized needs and evidence. | Candidate products, need IDs, quantities, rationale, confidence |
| Report assembly | Build modular report sections and client-visible methodology. | Qualified Opportunity Report, sources, warnings, appendix |
| Review & follow-up | Select evidence-backed actions, drill-downs, and optional QA supplements. | Actions, open questions, evidence review, final version |
Anatomy of a finding
A reviewable finding carries the source, applied context, calculation basis, uncertainty, and responsible action together so another person can inspect, challenge, and advance it.
Document, revision, page, sheet, section, table, detail, or highlighted region.
The requirement, condition, commercial signal, or quantity supported by that source.
Approved product, application, engineering, estimating, and qualification rules.
Scale, geometry, formula, inputs, units, allowances, cross-checks, and confidence.
Conflict, limitation, partial coverage, missing evidence, or UNKNOWN value.
Responsible role, question, approval, and next commercial or technical step.
Review-state dictionary
Buyers should know exactly how GAIA Civil labels ORION work that is incomplete, estimated, conflicted, or awaiting a person. Those states are part of the workflow contract-not edge-case UI.
The document or analysis has been admitted and work is not complete.
The lightweight lane produced a score, reasons, signals, and cited pages.
The finding has an addressable source path.
The configured verification requirements for that record were satisfied.
A bounded method produced a preliminary value with lower rigor.
A conflict, assumption, weak support, or decision remains unresolved.
The approved record cannot support the requested value.
Core work completed with missing or stubbed page coverage.
The document or analysis could not complete its required path.
Project, product, and engineering sources are being retrieved and connected to typed needs.
Scale candidates, dimensions, formulas, conflicts, and missing inputs are being recorded.
Weak evidence and decisions that can change the report are being routed for follow-up.
The dossier, approval state, owner, and selected integration or export path are being assembled.
Five transparency records
Methodology, calculation, evidence, version, and approval records answer different audit questions. Keeping them separate prevents a generic ‘audit trail’ label from hiding the actual detail.
Stages, methods, validations, measurements, assumptions, concise working summaries
Source, scale, geometry, formula, named inputs, units, allowances, confidence, conflicts
Document, revision, page, sheet/section/detail, region, category, evidence ID
Version number, parent/restored version, source message, timestamp, change summary
Role, open question, review state, export/submission decision, downstream handoff
Evaluation and implementation
A credible evaluation measures more than response quality. It verifies intake, page coverage, retrieval, calculations, citations, states, security, exports, and human approvals against a representative project.
Select representative plans, specs, addenda, schedules, and edge cases.
Load products, engineering rules, opportunity signals, exclusions, and accuracy profile.
Inspect ingestion coverage, retrieval, citations, calculations, conflicts, and report states.
Validate roles, security controls, approvals, exports, integrations, and acceptance criteria.
Workflow FAQ
The workflow is designed to make the path from source to action understandable at each state-not only after a polished report appears.
No. Direct intake can enter full analysis when the user already knows the project deserves it. Screening is the optional prioritization branch for larger candidate queues; relevant documents are promoted into the separate full-ingestion and analysis lifecycle.
The conflicting records remain addressable with their document/page context. Scale and quantity disagreement can cap confidence and generate a review flag. ORION can organize the conflict and draft a targeted clarification; the responsible reviewer resolves its technical or commercial effect.
Full ingestion renders every processed page, builds native page/image representations, and applies structured visual analysis. Valid component bounding boxes can create localized crops. If a visual pass fails, available text and raw page assets can remain as a marked stub; partial coverage is recorded.
QUOTE_READY is the strictest configured accuracy profile: per-opportunity verification, full BOM reconciliation, and rigorous engineering review. TAKEOFF runs broader independent scale and geometry analysis but may still surface unresolved questions. BALLPARK is optimized for rapid qualification.
No external write occurs without authorized review and approval. ORION can prepare API, webhook, MCP, JSON, CSV, PDF, takeoff, BOM, or inquiry artifacts inside GAIA Civil for approved integrations, but the final submission or downstream write remains governed.
Representative documents, product and engineering context, qualification and exclusion rules, target accuracy profile, responsible reviewers, data/security requirements, required outputs, and explicit acceptance criteria for citations, quantities, reports, and handoffs.
See GAIA Civil in context
Bring the source set, product context, estimating target, review roles, and outputs you want to validate.