> ## Documentation Index > Fetch the complete documentation index at: https://docs.trynia.ai/llms.txt > Use this file to discover all available pages before exploring further. # Data Extraction > Extract structured data from PDFs using JSON schemas, detect visual elements, and use engineering-specific extraction for technical documents Extract structured records from any PDF — financial filings, invoices, spec sheets, engineering drawings — using custom JSON schemas, visual element detection, or purpose-built engineering extraction. *** ## Three Extraction Modes Define a JSON schema describing the fields you need, and Nia extracts structured records from any PDF. Ideal for financial data, line items, tabular content, and repeating structures. Detect and locate visual elements — tables, figures, charts, diagrams — in PDF pages. Returns bounding boxes, classifications, and annotated page images. Purpose-built for technical documents — engineering drawings, P\&IDs, schematics, and spec sheets. Extracts structured metadata with optional follow-up queries for deeper analysis. *** ## How It Works Provide a PDF URL or an existing Nia source ID along with an optional page range. For table extraction, include a JSON schema defining the fields to extract. Nia parses the document, identifies relevant structures, and extracts data according to your schema (table mode) or built-in engineering models (engineering mode). Poll the extraction job until it completes. Table extraction returns an array of structured records; engineering extraction returns a result object you can query further. *** ## Table Extraction Define a JSON schema and Nia returns structured records matching your specification. This is ideal for pulling repeating data out of dense documents like SEC filings, invoices, or product catalogs. ### Start an Extraction Job ```bash theme={null} curl -X POST https://apigcp.trynia.ai/v2/extract \ -H "Authorization: Bearer $NIA_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "url": "https://www.sec.gov/Archives/edgar/data/1326801/000132680124000006/meta-20231231.htm", "page_range": "60-80", "json_schema": { "type": "object", "properties": { "line_item": { "type": "string", "description": "Name of the financial line item" }, "fiscal_year_2023": { "type": "number", "description": "Value for fiscal year 2023 in millions USD" }, "fiscal_year_2022": { "type": "number", "description": "Value for fiscal year 2022 in millions USD" }, "yoy_change_pct": { "type": "number", "description": "Year-over-year change as a percentage" } }, "required": ["line_item", "fiscal_year_2023"] } }' ``` Response: ```json theme={null} { "id": "ext_abc123", "status": "queued" } ``` ### Check Extraction Status ```bash theme={null} curl https://apigcp.trynia.ai/v2/extract/ext_abc123 \ -H "Authorization: Bearer $NIA_API_KEY" ``` Response when completed: ```json theme={null} { "id": "ext_abc123", "status": "completed", "progress": 100, "record_count": 24, "page_count": 20, "records": [ { "line_item": "Total revenue", "fiscal_year_2023": 134902, "fiscal_year_2022": 116609, "yoy_change_pct": 15.69 }, { "line_item": "Cost of revenue", "fiscal_year_2023": 38019, "fiscal_year_2022": 25249, "yoy_change_pct": 50.57 } ] } ``` ### JSON Schema Tips **Use descriptions** — Add a `description` to each field in your schema. Nia uses these to understand what data to look for, especially when column headers in the PDF are ambiguous. **Narrow the page range** — If you know which pages contain the data, specify `page_range` to speed up extraction and improve accuracy. *** ## Detect Extraction Detect and locate visual elements within PDF pages — tables, figures, charts, and diagrams. Detect mode returns bounding boxes and classifications for each element found, and can render annotated page images with the detections overlaid. ### Start a Detect Extraction Job ```bash theme={null} curl -X POST https://apigcp.trynia.ai/v2/extract/detect \ -H "Authorization: Bearer $NIA_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "url": "https://example.com/annual-report-2024.pdf", "page_range": "1-10", "include_symbols": false }' ``` | Parameter | Description | | ----------------- | ----------------------------------------------------------------------- | | `url` | URL of the PDF to process (provide either `url` or `source_id`) | | `source_id` | Source ID of an already-indexed document | | `page_range` | Pages to process (e.g. `"1-10"`, `"5,8,12"`) | | `include_symbols` | Enable symbol-level detection for technical documents (default `false`) | | `filter_pattern` | Regex to filter detected element types | Response: ```json theme={null} { "id": "det_abc123", "status": "queued", "type": "detect" } ``` ### Check Detect Extraction Status ```bash theme={null} curl https://apigcp.trynia.ai/v2/extract/detect/det_abc123 \ -H "Authorization: Bearer $NIA_API_KEY" ``` Response when completed: ```json theme={null} { "id": "det_abc123", "status": "completed", "progress": 100, "type": "detect", "page_count": 10, "result": { "pages": [ { "page_number": 1, "elements": [ { "type": "table", "bbox": [72, 200, 540, 450], "confidence": 0.97 }, { "type": "figure", "bbox": [72, 500, 400, 700], "confidence": 0.93 } ] } ] } } ``` ### Get Annotated Page Image Retrieve a page image with bounding boxes drawn over detected elements: ```bash theme={null} curl https://apigcp.trynia.ai/v2/extract/detect/det_abc123/page/1/image \ -H "Authorization: Bearer $NIA_API_KEY" \ --output page-1-annotated.png ``` This returns a PNG image with detection bounding boxes overlaid on the original page. **Use detect before table extraction** — Run detect first to identify which pages contain tables, then target those specific pages with table extraction for faster, more accurate results. *** ## Engineering Extraction Extract structured information from technical documents — engineering drawings, P\&IDs, schematics, datasheets, and construction specifications. Engineering mode uses specialized models tuned for technical content. ### Start an Engineering Extraction ```bash theme={null} curl -X POST https://apigcp.trynia.ai/v2/extract/engineering \ -H "Authorization: Bearer $NIA_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "url": "https://example.com/piping-diagram-rev3.pdf", "page_range": "1-5", "accuracy_mode": "precise" }' ``` The `accuracy_mode` parameter controls the speed/accuracy tradeoff: | Mode | Description | | --------- | ------------------------------------------------------------------------- | | `fast` | Optimized for speed. Good for initial scans and high-volume processing. | | `precise` | Maximum accuracy. Best for critical documents where every detail matters. | Response: ```json theme={null} { "id": "eng_xyz789", "status": "queued" } ``` ### Check Engineering Extraction Status ```bash theme={null} curl https://apigcp.trynia.ai/v2/extract/engineering/eng_xyz789 \ -H "Authorization: Bearer $NIA_API_KEY" ``` Response when completed: ```json theme={null} { "id": "eng_xyz789", "status": "completed", "result": { "document_type": "P&ID", "title": "Process Flow - Unit 400 Cooling System", "revision": "Rev 3", "components": [ { "tag": "P-401A", "type": "Centrifugal Pump", "specifications": "250 GPM, 150 PSI" }, { "tag": "HX-402", "type": "Shell and Tube Heat Exchanger", "specifications": "500 sq ft, 150 PSI design" } ] } } ``` ### Follow-Up Queries After an engineering extraction completes, you can ask follow-up questions about the results without re-processing the document: ```bash theme={null} curl -X POST https://apigcp.trynia.ai/v2/extract/engineering/eng_xyz789/query \ -H "Authorization: Bearer $NIA_API_KEY" \ -H "Content-Type: application/json" \ -d '{ "message": "What is the design pressure rating for all heat exchangers in this diagram?" }' ``` Response: ```json theme={null} { "id": "eng_xyz789", "chat_messages": [ { "role": "user", "content": "What is the design pressure rating for all heat exchangers in this diagram?" }, { "role": "assistant", "content": "Based on the extraction results, there is one heat exchanger in this diagram:\n\n- **HX-402** (Shell and Tube Heat Exchanger): Design pressure is **150 PSI**." } ] } ``` Follow-up queries use the already-extracted context, so they are fast and do not consume additional extraction credits. *** ## List All Extractions Retrieve all your extraction jobs, optionally filtered by type: ```bash theme={null} # List all extractions curl https://apigcp.trynia.ai/v2/extractions \ -H "Authorization: Bearer $NIA_API_KEY" # Filter by type curl "https://apigcp.trynia.ai/v2/extractions?type=table" \ -H "Authorization: Bearer $NIA_API_KEY" curl "https://apigcp.trynia.ai/v2/extractions?type=engineering" \ -H "Authorization: Bearer $NIA_API_KEY" ``` *** ## Extraction Statuses Both table and engineering extractions follow the same status lifecycle: | Status | Description | | ------------ | -------------------------------------------------------- | | `queued` | Job received and waiting to be processed | | `processing` | Extraction is actively running | | `completed` | Extraction finished successfully — results are available | | `failed` | Extraction encountered an error | *** ## Use Cases Extract line items, revenue figures, and balance sheet data from SEC filings (10-K, 10-Q) into structured records for analysis and comparison. Parse P\&IDs, wiring diagrams, and spec sheets to catalog components, materials, and specifications. Ask follow-up questions about extracted details. Pull vendor names, line items, quantities, and totals from invoices using a custom JSON schema tailored to your format. Extract equipment lists, compliance data, and specifications from engineering documents during M\&A or audits.