Files
eveAI/eveai_workers/processors/automagic_html_processor.py
Josako fda267b479 - Introduction of the Automatic HTML Processor
- Translation Service improvement
- Enable activation / deactivation of Processors
- Renew API-keys for Mistral (leading to workspaces)
- Align all Document views to use of a session catalog
- Allow for different processors for the same file type
2025-06-26 14:38:40 +02:00

66 lines
2.5 KiB
Python

import io
import pdfplumber
from flask import current_app
from langchain.text_splitter import RecursiveCharacterTextSplitter
from langchain_core.output_parsers import StrOutputParser
from langchain_core.prompts import ChatPromptTemplate
import re
from langchain_core.runnables import RunnablePassthrough
from common.eveai_model.tracked_mistral_ocr_client import TrackedMistralOcrClient
from common.extensions import minio_client
from common.utils.model_utils import create_language_template, get_embedding_llm, get_template
from .base_processor import BaseProcessor
from common.utils.business_event_context import current_event
from .processor_registry import ProcessorRegistry
class AutomagicHTMLProcessor(BaseProcessor):
def __init__(self, tenant, document_version, catalog, processor):
super().__init__(tenant, document_version, catalog, processor)
self.chunk_size = catalog.max_chunk_size
self.chunk_overlap = 0
self.tuning = self.processor.tuning
self.prompt_params = {
"custom_instructions": self.processor.configuration.get("custom_instructions", ""),
}
template, llm = get_template("automagic_html_parse")
translation_prompt = ChatPromptTemplate.from_template(template)
setup = RunnablePassthrough()
output_parser = StrOutputParser()
self.chain = (setup | translation_prompt | llm | output_parser)
def process(self):
self._log("Starting Automagic HTML processing")
try:
# Get HTML-file data
file_data = minio_client.download_document_file(
self.tenant.id,
self.document_version.bucket_name,
self.document_version.object_name,
)
# Invoke HTML Processing Agent
self.prompt_params["html"] = file_data
with current_event.create_span("Markdown Generation"):
markdown = self.chain.invoke(self.prompt_params)
self._save_markdown(markdown)
# Retrieve Title
match = re.search(r'^# (.+)', markdown, re.MULTILINE)
title = match.group(1).strip() if match else None
self._log("Finished Automagic HTML Processing")
return markdown, title
except Exception as e:
self._log(f"Error automagically processing HTML: {str(e)}", level='error')
raise
# Register the processor
ProcessorRegistry.register("AUTOMAGIC_HTML_PROCESSOR", AutomagicHTMLProcessor)