AI Enhancements Roadmap (P5)

Note: This plan is a working draft. Some technical details (exact file provenance, build parity, vendor timelines, pricing, and Autodesk product capabilities) may need verification as we inventory the full legacy tree and validate behavior in modern environments.

Goal

Add AI-assisted workflows that reduce manual paperwork and generate “panel book” artifacts faster, while staying audit-friendly and predictable.

Key framing:

  • AI is an assistant, not the authority.

  • Human review is required for anything that can affect deliverables.

P5 capabilities (incremental)

1) Auto-fill forms (assist mode)

Inputs:

  • drawing metadata

  • project templates

  • user notes

  • structured outputs from ConstructiVision run

Outputs:

  • pre-filled complaint / submittal / job forms

  • highlighted uncertainties for human review

Success metric:

  • 50—80% of fields auto-filled with low correction burden

2) “EZ Button” panel book generator

A guided workflow that:

  • collects required inputs

  • validates completeness

  • runs generation

  • produces a final package (PDF + supporting artifacts)

Success metric:

  • panel book generated in minutes with consistent formatting

3) Integrated tests + evidence packs

Automatically generates:

  • test results summary

  • traceability links (inputs → computations → outputs)

  • “evidence pack” bundle for audit/review

Success metric:

  • evidence pack created every run with no extra work

Technical approach (safe-by-design)

  • Use structured intermediate data formats (CSV/JSON) as AI inputs

  • Keep deterministic rules for:

    • calculations

    • formatting

    • acceptance checks

  • AI focuses on:

    • extraction

    • suggestion

    • summarization

    • drafting

Parallelization plan

Ideal path (if web dev hired during P4):

  • Build the cloud artifact pipeline first

  • Add AI services as optional steps in that pipeline

  • Keep local plugin working even if AI is off

Fallback path (no staffing):

  • Prototype AI workflows after P4 stabilizes

  • Start with a small set of forms and one “EZ Button” workflow

Risks to manage

  • hallucinated values (must never silently enter deliverables)

  • privacy / customer data handling

  • inconsistent outputs between versions

  • user trust: AI must be transparent and reversible

Deliverables

  • AI requirements doc (inputs/outputs, review gates)

  • Prototype UI (review/approve)

  • Audit-friendly logging and traceability