915f827305
## ✨ Novedades - **Soporte LaTeX**: Generación de PDFs y DOCX con fórmulas matemáticas renderizadas correctamente usando Pandoc. - **Sanitización Automática**: Corrección de caracteres Unicode (griegos/cirílicos) y sintaxis LaTeX para evitar errores de compilación. - **GLM/Claude Prioritario**: Cambio de proveedor de IA predeterminado a Claude/GLM para mayor estabilidad y capacidad de razonamiento. - **Mejoras en Formato**: El formateo final del resumen ahora usa el modelo principal (GLM) en lugar de Gemini para consistencia. ## 🛠️ Cambios Técnicos - `document/generators.py`: Reemplazo de generación manual por `pandoc`. Añadida función `_sanitize_latex`. - `services/ai/claude_provider.py`: Soporte mejorado para variables de entorno de Z.ai. - `services/ai/provider_factory.py`: Prioridad ajustada `Claude > Gemini`. - `latex/`: Añadida documentación de referencia para el pipeline LaTeX.
353 lines
13 KiB
Python
353 lines
13 KiB
Python
"""
|
||
Document generation utilities
|
||
"""
|
||
|
||
import logging
|
||
import subprocess
|
||
import re
|
||
from pathlib import Path
|
||
from typing import Dict, Any, List, Tuple
|
||
from core import FileProcessingError
|
||
from config import settings
|
||
from services.ai import ai_provider_factory
|
||
|
||
|
||
class DocumentGenerator:
|
||
"""Generate documents from processed text"""
|
||
|
||
def __init__(self):
|
||
self.logger = logging.getLogger(__name__)
|
||
self.ai_provider = ai_provider_factory.get_best_provider()
|
||
|
||
def generate_summary(
|
||
self, text: str, base_name: str
|
||
) -> Tuple[bool, str, Dict[str, Any]]:
|
||
"""Generate unified summary"""
|
||
self.logger.info(f"Generating summary for {base_name}")
|
||
|
||
try:
|
||
# Step 1: Generate Bullet Points (Chunking handled by provider or single prompt for now)
|
||
# Note: We use the main provider (Claude/Zai) for content generation
|
||
self.logger.info("Generating bullet points...")
|
||
bullet_prompt = f"""Analiza el siguiente texto y extrae entre 5 y 8 bullet points clave en español.
|
||
|
||
REGLAS ESTRICTAS:
|
||
1. Devuelve ÚNICAMENTE bullet points, cada línea iniciando con "- "
|
||
2. Cada bullet debe ser conciso (12-20 palabras) y resaltar datos, fechas, conceptos o conclusiones importantes
|
||
3. NO agregues introducciones, conclusiones ni texto explicativo
|
||
4. Concéntrate en los puntos más importantes del texto
|
||
5. Incluye fechas, datos específicos y nombres relevantes si los hay
|
||
|
||
Texto:
|
||
{text[:15000]}""" # Truncate to avoid context limits if necessary, though providers handle it differently
|
||
|
||
try:
|
||
bullet_points = self.ai_provider.generate_text(bullet_prompt)
|
||
self.logger.info(f"Bullet points generated: {len(bullet_points)}")
|
||
except Exception as e:
|
||
self.logger.warning(f"Bullet point generation failed: {e}")
|
||
bullet_points = "- Puntos clave no disponibles por error en IA"
|
||
|
||
# Step 2: Generate Unified Summary
|
||
self.logger.info("Generating unified summary...")
|
||
summary_prompt = f"""Eres un profesor universitario experto en historia y economía. Redacta un resumen académico integrado en español usando el texto y los bullet points extraídos.
|
||
|
||
REQUISITOS ESTRICTOS DE CONTENIDO:
|
||
- Extensión entre 500-700 palabras
|
||
- Usa encabezados Markdown con jerarquía clara (##, ###)
|
||
- Desarrolla los puntos clave con profundidad y contexto histórico/económico
|
||
- Mantén un tono académico y analítico
|
||
- Incluye conclusiones significativas
|
||
- NO agregues texto fuera del resumen
|
||
- Devuelve únicamente el resumen en formato Markdown
|
||
|
||
REQUISITOS ESTRICTOS DE FORMATO MATEMÁTICO (LaTeX):
|
||
- Si el texto incluye fórmulas matemáticas o económicas, DEBES usar formato LaTeX.
|
||
- Usa bloques $$ ... $$ para ecuaciones centradas importantes.
|
||
- Usa $ ... $ para ecuaciones en línea.
|
||
- Ejemplo: La fórmula del interés compuesto es $A = P(1 + r/n)^{{nt}}$.
|
||
- NO uses bloques de código (```latex) para las fórmulas, úsalas directamente en el texto para que Pandoc las renderice.
|
||
|
||
Contenido a resumir:
|
||
{text[:20000]}
|
||
|
||
Puntos clave a incluir obligatoriamente:
|
||
{bullet_points}"""
|
||
|
||
try:
|
||
raw_summary = self.ai_provider.generate_text(summary_prompt)
|
||
except Exception as e:
|
||
self.logger.error(f"Raw summary generation failed: {e}")
|
||
raise e
|
||
|
||
# Step 3: Format with IA (using main provider instead of Gemini)
|
||
self.logger.info("Formatting summary with IA...")
|
||
format_prompt = f"""Revisa y mejora el siguiente resumen en Markdown para que sea perfectamente legible y compatible con Pandoc:
|
||
|
||
{raw_summary}
|
||
|
||
Instrucciones:
|
||
- Corrige cualquier error de formato Markdown
|
||
- Asegúrate de que los encabezados estén bien espaciados
|
||
- Verifica que las viñetas usen "- " correctamente
|
||
- Mantén exactamente el contenido existente
|
||
- EVITA el uso excesivo de negritas (asteriscos), úsalas solo para conceptos clave
|
||
- VERIFICA que todas las fórmulas matemáticas estén correctamente encerradas en $...$ (inline) o $$...$$ (display)
|
||
- NO alteres la sintaxis LaTeX dentro de los delimitadores $...$ o $$...$$
|
||
- Devuelve únicamente el resumen formateado sin texto adicional"""
|
||
|
||
try:
|
||
# Use the main provider (Claude/GLM) for formatting too
|
||
if self.ai_provider.is_available():
|
||
summary = self.ai_provider.generate_text(format_prompt)
|
||
else:
|
||
self.logger.warning(
|
||
"AI provider not available for formatting, using raw summary"
|
||
)
|
||
summary = raw_summary
|
||
except Exception as e:
|
||
self.logger.warning(f"Formatting failed ({e}), using raw summary")
|
||
summary = raw_summary
|
||
|
||
# Generate filename
|
||
filename = self._generate_filename(text, summary)
|
||
|
||
# Create document
|
||
markdown_path = self._create_markdown(summary, base_name)
|
||
|
||
docx_path = None
|
||
try:
|
||
docx_path = self._create_docx(markdown_path, base_name)
|
||
except Exception as e:
|
||
self.logger.error(f"Failed to create DOCX (non-critical): {e}")
|
||
|
||
pdf_path = None
|
||
try:
|
||
# Sanitize LaTeX before PDF generation
|
||
self._sanitize_latex(markdown_path)
|
||
pdf_path = self._create_pdf(markdown_path, base_name)
|
||
except Exception as e:
|
||
self.logger.error(f"Failed to create PDF (non-critical): {e}")
|
||
|
||
# Upload to Notion if configured
|
||
from services.notion_service import notion_service
|
||
|
||
notion_uploaded = False
|
||
notion_page_id = None
|
||
if settings.has_notion_config:
|
||
try:
|
||
title = base_name.replace("_", " ").title()
|
||
|
||
# Crear página con el contenido completo del resumen
|
||
notion_metadata = {
|
||
"file_type": "Audio", # O 'PDF' dependiendo del origen
|
||
"pdf_path": pdf_path if pdf_path else Path(""),
|
||
"add_status": False, # No usar Status/Tipo (no existen en la DB)
|
||
"use_as_page": False, # Usar como database, no página
|
||
}
|
||
|
||
notion_page_id = notion_service.create_page_with_summary(
|
||
title=title, summary=summary, metadata=notion_metadata
|
||
)
|
||
|
||
if notion_page_id:
|
||
notion_uploaded = True
|
||
self.logger.info(
|
||
f"✅ Resumen subido a Notion: {title} (ID: {notion_page_id})"
|
||
)
|
||
else:
|
||
self.logger.warning(f"⚠️ No se pudo subir a Notion: {title}")
|
||
except Exception as e:
|
||
self.logger.warning(f"❌ Error al subir a Notion: {e}")
|
||
import traceback
|
||
|
||
traceback.print_exc()
|
||
else:
|
||
self.logger.info("Notion not configured - skipping upload")
|
||
|
||
metadata = {
|
||
"markdown_path": str(markdown_path),
|
||
"docx_path": str(docx_path) if docx_path else "",
|
||
"pdf_path": str(pdf_path) if pdf_path else "",
|
||
"docx_name": Path(docx_path).name if docx_path else "",
|
||
"summary": summary,
|
||
"filename": filename,
|
||
"notion_uploaded": notion_uploaded,
|
||
"notion_page_id": notion_page_id,
|
||
}
|
||
|
||
return True, summary, metadata
|
||
|
||
except Exception as e:
|
||
self.logger.error(f"Document generation process failed: {e}")
|
||
return False, "", {}
|
||
|
||
def _sanitize_latex(self, markdown_path: Path) -> None:
|
||
"""Sanitize LaTeX syntax in Markdown file to prevent Pandoc errors"""
|
||
try:
|
||
content = markdown_path.read_text(encoding="utf-8")
|
||
|
||
# 1. Unescape escaped dollar signs which are common LLM errors for math
|
||
content = content.replace(r"\$", "$")
|
||
|
||
# 2. Fix common Cyrillic and Greek characters that sneak in via LLMs
|
||
replacements = {
|
||
"ч": "ch",
|
||
"в": "v",
|
||
"к": "k",
|
||
"м": "m",
|
||
"н": "n",
|
||
"т": "t",
|
||
"—": "-",
|
||
"–": "-",
|
||
"“": '"',
|
||
"”": '"',
|
||
"’": "'",
|
||
"Δ": "$\\Delta$",
|
||
"δ": "$\\delta$",
|
||
"Σ": "$\\Sigma$",
|
||
"σ": "$\\sigma$",
|
||
"π": "$\\pi$",
|
||
"Π": "$\\Pi$",
|
||
"α": "$\\alpha$",
|
||
"β": "$\\beta$",
|
||
"γ": "$\\gamma$",
|
||
"θ": "$\\theta$",
|
||
"λ": "$\\lambda$",
|
||
"μ": "$\\mu$",
|
||
}
|
||
|
||
# Be careful not to double-replace already correct LaTeX
|
||
for char, repl in replacements.items():
|
||
if char in content:
|
||
# Check if it's already inside math mode would be complex,
|
||
# but for now we assume raw unicode greek chars should become latex
|
||
content = content.replace(char, repl)
|
||
|
||
markdown_path.write_text(content, encoding="utf-8")
|
||
self.logger.info(f"Sanitized LaTeX in {markdown_path}")
|
||
except Exception as e:
|
||
self.logger.warning(f"Failed to sanitize LaTeX: {e}")
|
||
|
||
def _generate_filename(self, text: str, summary: str) -> str:
|
||
"""Generate intelligent filename"""
|
||
try:
|
||
# Use AI to extract key topics
|
||
prompt = f"""Extract 2-3 key topics from this summary to create a filename.
|
||
Summary: {summary}
|
||
|
||
Return only the topics separated by hyphens, max 20 chars each, in Spanish:"""
|
||
|
||
try:
|
||
topics_text = self.ai_provider.generate_text(prompt)
|
||
except Exception:
|
||
topics_text = summary[:100]
|
||
|
||
# Simple topic extraction
|
||
topics = re.findall(r"\b[A-ZÁÉÍÓÚÑ][a-záéíóúñ]+\b", topics_text)[:3]
|
||
if not topics:
|
||
topics = ["documento"]
|
||
|
||
# Limit topic length
|
||
topics = [t[: settings.MAX_FILENAME_TOPICS_LENGTH] for t in topics]
|
||
|
||
filename = "_".join(topics)[: settings.MAX_FILENAME_LENGTH]
|
||
return filename
|
||
|
||
except Exception as e:
|
||
self.logger.error(f"Filename generation failed: {e}")
|
||
return "documento"
|
||
|
||
def _create_markdown(self, summary: str, base_name: str) -> Path:
|
||
"""Create Markdown document"""
|
||
output_dir = settings.LOCAL_DOWNLOADS_PATH
|
||
output_dir.mkdir(parents=True, exist_ok=True)
|
||
|
||
output_path = output_dir / f"{base_name}_unificado.md"
|
||
|
||
content = f"""# {base_name.replace("_", " ").title()}
|
||
|
||
## Resumen
|
||
|
||
{summary}
|
||
|
||
---
|
||
|
||
*Generado por CBCFacil*
|
||
"""
|
||
|
||
with open(output_path, "w", encoding="utf-8") as f:
|
||
f.write(content)
|
||
|
||
return output_path
|
||
|
||
def _create_docx(self, markdown_path: Path, base_name: str) -> Path:
|
||
"""Create DOCX document using pandoc"""
|
||
output_dir = settings.LOCAL_DOCX
|
||
output_dir.mkdir(parents=True, exist_ok=True)
|
||
|
||
output_path = output_dir / f"{base_name}_unificado.docx"
|
||
|
||
self.logger.info(
|
||
f"Converting Markdown to DOCX: {markdown_path} -> {output_path}"
|
||
)
|
||
|
||
try:
|
||
cmd = [
|
||
"pandoc",
|
||
str(markdown_path),
|
||
"-o",
|
||
str(output_path),
|
||
"--from=markdown",
|
||
"--to=docx",
|
||
]
|
||
|
||
result = subprocess.run(cmd, capture_output=True, text=True, check=True)
|
||
|
||
self.logger.info("DOCX generated successfully with pandoc")
|
||
return output_path
|
||
|
||
except subprocess.CalledProcessError as e:
|
||
self.logger.error(f"Pandoc DOCX conversion failed: {e.stderr}")
|
||
raise FileProcessingError(f"Failed to generate DOCX: {e.stderr}")
|
||
except Exception as e:
|
||
self.logger.error(f"Error generating DOCX: {e}")
|
||
raise FileProcessingError(f"Error generating DOCX: {e}")
|
||
|
||
def _create_pdf(self, markdown_path: Path, base_name: str) -> Path:
|
||
"""Create PDF document using pandoc and pdflatex"""
|
||
output_dir = settings.LOCAL_DOWNLOADS_PATH
|
||
output_dir.mkdir(parents=True, exist_ok=True)
|
||
|
||
output_path = output_dir / f"{base_name}_unificado.pdf"
|
||
|
||
self.logger.info(
|
||
f"Converting Markdown to PDF: {markdown_path} -> {output_path}"
|
||
)
|
||
|
||
try:
|
||
cmd = [
|
||
"pandoc",
|
||
str(markdown_path),
|
||
"-o",
|
||
str(output_path),
|
||
"--pdf-engine=pdflatex",
|
||
"-V",
|
||
"geometry:margin=2.5cm",
|
||
"-V",
|
||
"fontsize=12pt",
|
||
"--highlight-style=tango",
|
||
]
|
||
|
||
result = subprocess.run(cmd, capture_output=True, text=True, check=True)
|
||
|
||
self.logger.info("PDF generated successfully with pandoc")
|
||
return output_path
|
||
|
||
except subprocess.CalledProcessError as e:
|
||
self.logger.error(f"Pandoc PDF conversion failed: {e.stderr}")
|
||
raise FileProcessingError(f"Failed to generate PDF: {e.stderr}")
|
||
except Exception as e:
|
||
self.logger.error(f"Error generating PDF: {e}")
|
||
raise FileProcessingError(f"Error generating PDF: {e}")
|