Admin (web) → processUploadedFile (ORPC, admin)
                ↓
          regular-processing.ts
                ↓
    documentProcessor.processDocument()
                ↓
    POST http://DOCUMENT_PROCESSOR_URL/process
                ↓
    document-processor microservice (FastAPI)
                ↓
    Chunks → INSERT INTO chunks (DB)

# Variables d'environnement
DOCUMENT_PROCESSOR_HOST=0.0.0.0
DOCUMENT_PROCESSOR_PORT=8001
MAX_FILE_SIZE_MB=50
ENABLE_OCR=false
TESSERACT_PATH=         # optionnel, si tesseract n'est pas dans PATH
LOG_LEVEL=INFO

cd apps/document-processor
pip install -e .
uvicorn src.main:app --host 0.0.0.0 --port 8001

{
  "status": "healthy",
  "timestamp": "2024-01-15T10:30:00.000Z"
}

{
  file_content: string;        // base64, data URL, ou texte brut (pour text/plain, text/markdown, text/csv, application/json)
  file_name: string;
  file_type: string;           // MIME type (voir formats supportés)
  chunk_size?: number;         // default: 1000 (taille en caractères)
  overlap?: number;            // default: 200 (chevauchement entre chunks)
  extract_qa_pairs?: boolean;  // default: false
  extract_tables?: boolean;    // default: true
  extract_images?: boolean;    // default: false
}

{
  success: boolean;
  chunks: Chunk[];           // liste des chunks extraits
  qa_pairs: QAPair[];        // si extract_qa_pairs: true
  tables: Table[];           // tableaux extraits (si extract_tables: true)
  images: Image[];           // images extraites (si extract_images: true)
  metadata: Record<string, any>;
  processing_time_ms: number;
  error?: string;            // présent si success: false
}

type Chunk = {
  rank: number;
  content: string;
  category: string; // "not categorized" par défaut
  metadata?: {
    source_type: string;      // "text", "table", "picture"
    section_path?: string[];  // hiérarchie des sections
    page_numbers?: number[];  // pages du chunk
    bboxes?: Array<{l, t, r, b}>;  // coordonnées
    docling_item_refs?: string[];  // références items Docling
    token_count?: number;     // nombre de tokens
    origin?: string;          // nom du fichier d'origine
    parser_version: string;
    chunking_config: { max_tokens: number, merge_peers: boolean };
  };
  page_number?: number;
  section_title?: string;
  section_path?: string[];
  provenance?: {
    page_numbers?: number[];
    bboxes?: Array<{l, t, r, b}>;
    docling_item_refs?: string[];
  };
  token_count?: number;
  source_type?: string;
  has_table?: boolean;
  has_image?: boolean;
}

DOCLING_ENABLE_OCR=true
DOCLING_OCR_ENGINE=easyocr  # ou rapidocr, tesseract-cli, tesserocr
DOCLING_OCR_LANGUAGES=en,fr

documentProcessor.processDocument(request: {
  file_content: string,
  file_name: string,
  file_type: string,
  chunk_size: number,
  overlap: number,
  extract_qa_pairs: boolean,
  extract_tables: boolean,
  extract_images: boolean
}): Promise<ProcessingResult>