feat: mid-sized structured asset creator workflow

[droarty]

Plan: Mid-Sized Structured Asset Creator — Configurable Workflow

Build a generic mid-sized-structured-asset-creator workflow. Users upload a PDF, then engage in a multi-turn chat where Claude: (1) analyzes the document structure, (2) collects an asset name and category, (3) proposes and refines a schema through conversation, and (4) extracts all records once the schema is confirmed. Both the schema and data are persisted together in a StructuredAsset MongoDB model.

Architecture

Phase 1 — Upload

• HTTP POST /api/documents/:id/upload (multer)
• Files API: upload PDF → store fileId in state
• AI file-chat: analyze structure, ask for name + category, propose schema

Phase 2 — Schema discussion (multi-turn chat)

• New file-chat AI step type in workflow engine
• Each user message continues the conversation with the PDF in context
• Claude proposes/refines schema, triggers schema-proposed response type

Phase 2b — Version check (immediately after schema-proposed)

• schema-proposed handler step 2: check-asset-exists query
• asset-not-found → saveIntent: 'create', show "Confirm Schema"
• asset-exists → uploadStatus: 'version-check', show 3-choice conflict card:
  • Replace v{N}
  • Save as v{N+1}
  • Change name / category (re-opens chat, loops back)

Phase 3 — Extraction

• New file-extract AI step type
• One-shot extraction using confirmed schema

Phase 4 — Accept / Reject

• accept-import → persist-structured-asset (reads saveIntent — no second conflict check needed)
• reject-import → back to schema-proposed phase

New Files

• libs/shared-types/src/structured-asset.types.ts
• apps/api/src/app/models/structured-asset.model.ts
• apps/api/src/app/services/fileExtract.service.ts
• apps/api/src/app/routes/upload.routes.ts
• apps/api/src/app/config/workflows/mid-sized-structured-asset-creator.json
• apps/web/src/app/components/ChatUploadPanel.tsx
• apps/web/src/app/components/JsonViewer.tsx

Modified Files

• WorkflowEngine.ts — extend AiStepConfig with type, fileIdPath, historyPath, userMessagePath, schemaPath
• AIEventManager.ts — route file-chat / file-extract to new service methods
• EventProcessorWorker.ts — resolve file context and pass to AI manager
• QueryExecutor.ts — add check-asset-exists and persist-structured-asset queries
• document.model.ts — add mid-sized-structured-asset-creator to enum
• layoutRegistry.ts — register chatUploadPanel and jsonViewer
• app.ts — register upload route

Data Model: StructuredAsset

• name, category, assetVersion (compound unique index)
• schema (embedded JSON Schema + description)
• data (array of extracted records)
• recordCount, sourceFileName, sourceDocumentId, timestamps

Dependencies

• multer + @types/multer (new)
• @anthropic-ai/sdk already installed