README.md

PDF Processing

This folder contains the PDF-native and validated-text fallback workflow used by CV optimization.

Main entry points

• pdf-native-optimization-workflow.ts: native-first orchestration with allowlisted fallback to extracted text.
• pdf-text-extraction.ts: page-by-page extraction and reconstruction using pdfjs-dist.
• pdf-text-quality.ts: fail-closed quality heuristics for extracted text.
• pdf-native-provider-fallback.ts: classifies provider failures to decide if text fallback is allowed.

Data flow

flowchart TD
    A["Sanitized PDF upload"] --> B["Native provider PDF attempt"]
    B -->|"success"| C["Optimized HTML"]
    B -->|"explicit files/PDF failure"| D["Page-by-page text extraction"]
    D --> E["Quality heuristics"]
    E -->|"pass"| F["Same provider/model with extracted text"]
    E -->|"fail"| G["Explicit user-facing extraction error"]
    F --> C

Loading

Important behavior

• Native PDF input is always attempted first.
• Text fallback is allowed only for explicit file/PDF capability, file-credits, or file-endpoint failures.
• Generic provider failures do not trigger fallback.
• If extraction quality is suspicious, the workflow fails with an explicit no-OCR message instead of sending incomplete text to the LLM.