PDF Parsing & Document AI Expert Needed to Improve an Existing System

The system should accurately understand complex page structures, including multi-column layouts, mixed content, tables, captions, images, headers, footnotes, and other document regions. Table and glyph extraction Tables must be extracted while preserving: 1. Merged cells 2. Header relationships 3. Reading order 4. Special glyphs and symbols such as e.g ✓, ⊗, ●, and ○ 4. Technical image and diagram extraction The system must correctly detect and crop complete: 1. Technical drawings 2. Process-flow diagrams 3. Architecture diagrams 4. Schematics and other embedded figures Images and diagrams must not be cropped in the middle or divided incorrectly across multiple regions. Responsibilities - Review our existing codebase and parsing pipeline - Evaluate current extraction quality and identify failure patterns - Improve the existing architecture build around and model-routing logic - Determine which parser or extraction method should be used for each content type - Implement validation, confidence scoring, and fallback mechanisms - Improve consistency across different PDF layouts and versions - Test improvements against a representative set of documents - Define measurable accuracy and quality metrics - Document the changes and technical decisions We are open/looking for a hybrid or agentic approach that combines native PDF extraction, OCR, layout-detection models, multimodal models, and specialised parsing tools. Please apply only if you have previously built or improved a similar production-level PDF parsing or Document AI system.