> ## Documentation Index > Fetch the complete documentation index at: https://help.1nspect.app/llms.txt > Use this file to discover all available pages before exploring further. # Schema Upload & Review # Schema Upload & Review When you want to build a template from an **existing PDF report** (e.g. a competitor's template, a state-mandated form, or your own legacy report) instead of building from scratch, the Schema Upload & Review flow is the fast path. The AI reads the PDF, extracts sections / subsections / fields / checkboxes, and presents them for your review before saving. This article documents both halves of the flow: * **Schema Upload** (`/templates/builder/:templateId/schema`) — submit a PDF, watch the AI extract * **Schema Review** (`/template-review` or `/template-editor/:id`) — review the extraction and save into a template *** ## When to use this flow * **Building a template from a competitor's PDF** — drop in a copy of their report, get most of the structure for free * **Cloning a TREC-variant report** another inspector uses * **Migrating from a legacy report format** to 1nspecT-rendered reports * **Starting from a state-mandated form** (e.g. some commercial standards) Use the [Template Editor](/1nspect/admin-operator/templates/template-editor) or the deeper schema editor for hand-building from scratch. Use this flow when an existing PDF is your starting point. *** ## Phase 1 — Schema Upload \[SCREENSHOT: schema-upload-screen.png — desktop, the Schema Upload page showing a large drag-and-drop zone for PDF upload, with progress indicators for extraction phases below.] ### Uploading a PDF The upload zone accepts: * **PDF** files (any size up to your tenant's configured limit, typically 25 MB) * DOCX files (extracted via the same pipeline) Drag a file onto the zone or click to browse. The file uploads, and the extraction pipeline runs. ### Extraction phases Behind the scenes, the AI runs in several phases visible as progress indicators: | Phase | What happens | | ------------------------- | ---------------------------------------------------------------------------------------- | | **OCR** | If the PDF has image-only pages, OCR converts them to searchable text | | **Structural extraction** | The AI identifies sections, subsections, headings, and hierarchy | | **Field extraction** | Text inputs, checkboxes, rating columns, and informational fields are identified by type | | **Validation** | The extracted structure is checked for invariants (unique IDs, valid field types) | Total time: typically 30–90 seconds depending on PDF complexity. A long PDF (50+ pages) can take 2–3 minutes. ### Cost Schema extraction uses the `extract-structured-ocr` AI operation: **3 IT** per extraction (see [AI Features → Complete Operation Reference](/1nspect/features/ai-features#complete-ai-operation-cost-reference)). This is per-attempt — re-running on the same PDF costs additional IT. ### What gets extracted The pipeline produces a draft schema with: * **Sections** with names and short codes * **Subsections** under each section * **Fields** of various types, classified automatically: | Field type | Recognized from | | ---------------------- | ------------------------------------------------------- | | **text** | Free-form text input lines | | **checkbox** | Single-state checkboxes | | **informational** | Labeled data-capture fields (e.g. "Year Built", "Sqft") | | **rating** | I/NI/NP/D-style rating columns or pass/fail boxes | | **comment** | Free-form comment/narrative regions | | **photo\_placeholder** | Image-placeholder boxes | Each field is also tagged with a **field category** — `standard` or `informational`. This drives whether the field appears in the mobile inspector's main flow or in the tech-spec grid. *** ## Phase 2 — Schema Review After extraction completes, you are auto-routed to the **Review Template Schema** page. URL: `/template-review` or `/template-editor/:id`. \[SCREENSHOT: schema-review-screen.png — desktop, the Schema Review page showing Template Information at the top, an Extraction Summary card with counts, and the extracted Template Structure tree below.] ### Page sections #### Template Information | Field | Notes | | ------------- | --------------------------------------------------------------------------------------- | | Template Name | Pre-filled from the PDF title; editable | | Standard | Drop-down: **TREC** (Texas) · **ASHI** · **InterNACHI** · **CCPIA ComSOP** · **Custom** | | Description | Optional internal notes | Choose the standard that best matches the PDF you uploaded. This affects how the AI tunes narrative language later when inspectors use the template. #### Extraction Summary A card showing what the AI found: | Stat | Description | | ---------- | ---------------------------- | | Sections | Top-level sections detected | | Checkboxes | Individual checkbox elements | | Text Boxes | Text-input fields | | Paragraphs | Narrative/comment regions | The numbers help you spot extraction issues at a glance. If a 30-page report extracts as 2 sections, something went wrong — likely the AI didn't recognize the section breaks. Re-upload with a cleaner version, or hand-edit the extracted schema. #### Template Structure A hierarchical tree showing the extracted schema. Each node: * **Section** (collapsible) — name + count of subsections * **Subsection** (collapsible) — name + count of fields * **Field** — type + category Click any node to: * **Edit** — change name, type, category * **Delete** — remove this node from the schema * **Add child** — insert a new section, subsection, or field The Review page is **fully editable** — fix anything the AI got wrong before saving. ### Common AI mistakes to look for Based on patterns the extraction pipeline has known issues with: | Mistake | How to fix | | ----------------------------- | ----------------------------------------------------------------------------------------------------------------------- | | **Missing sections** | Bigger headings sometimes get classified as subsections. Drag-to-promote in the tree, or manually add a parent section. | | **Misclassified field types** | A rating column sometimes extracts as text. Edit and change the type. | | **Duplicate fields** | Multi-column layouts can extract the same field twice. Delete duplicates. | | **Missing fields** | Sparse PDFs (lots of whitespace, image-only layouts) may miss fields entirely. Manually add. | | **Wrong section order** | The hierarchy preserves source order; if the PDF was unusually arranged, drag to reorder. | *** ## Saving the reviewed schema A **Save** button at the bottom (or top-right) commits the schema: | What happens on Save | | ----------------------------------------------------------------------------------------------------- | | The schema replaces the current template's schema (if editing an existing template) | | If this was a fresh extraction with no template selected, a new template is created using this schema | | The template appears in [Template Builder](/1nspect/admin-operator/templates/template-builder) | After save, you typically continue to [Template Designer](/1nspect/admin-operator/templates/template-designer) to set up the canvas and rating scale. *** ## Re-extracting If you want to start the AI extraction over (e.g. after editing the source PDF): 1. Return to the Upload page 2. Drop the new PDF 3. The pipeline re-runs and replaces the current extraction Re-extraction **discards** any hand-edits you made to the previous Review session. Save before re-extracting if you want to preserve those edits in the template. *** ## Errors and recovery | Alert | Cause | Fix | | ----------------------------------- | ------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------- | | **PDF upload failed** | Network or file-size issue | Check the file size; try a smaller version | | **Extraction failed** | AI service error, malformed PDF, or scan quality too low | Re-upload a higher-quality PDF, or build manually in [Template Editor](/1nspect/admin-operator/templates/template-editor) | | **No sections detected** | PDF lacks clear section structure (e.g. one giant text block) | Hand-add sections in the Review page, or split the PDF into per-section files and extract each | | **Validation error: duplicate IDs** | The AI generated overlapping field IDs (rare) | Save will be blocked until duplicates are resolved — the Review page surfaces specific items to fix | | **Save failed** | Network or auth error | Retry; if persistent, your Review edits are preserved in browser storage and recovered on refresh | *** ## What this flow does NOT do * **It does not design the report PDF's appearance.** Schema is structure only; rendering is [Template Designer](/1nspect/admin-operator/templates/template-designer). * **It does not assign default narratives or AI prompts.** Those are configured separately via [Narrative Library](/1nspect/admin-operator/intelligence-hub/narrative-library). * **It does not validate that the extracted schema matches an official standard's spec.** The AI extracts whatever's in the PDF; if the PDF is non-compliant, the extraction reflects that. *** ## Related articles * [Templates Overview](/1nspect/admin-operator/templates/overview) * [Template Builder](/1nspect/admin-operator/templates/template-builder) — where the newly-extracted template appears * [Template Designer](/1nspect/admin-operator/templates/template-designer) — set up the canvas after schema is in place * [Template Editor](/1nspect/admin-operator/templates/template-editor) — edit settings, font, branding * [AI Features → Schema extraction](/1nspect/features/ai-features) — IT cost and operation reference