from datetime import datetime import re from langchain_core.messages import HumanMessage, SystemMessage from langgraph.constants import Send from LLMS import get_llm, get_planner from LLMS.models import Section, Subsection from state import BlogState, WorkerState def orchestrator(state: BlogState): """Orchestrator that generates a plan for the blog based on news items""" try: # Parse the content to extract news items content_lines = state['content'].split('\n\n') news_items = [] current_item = {} for content_block in content_lines: if content_block.startswith('TITLE:'): # Start of a new item if current_item and 'title' in current_item: news_items.append(current_item) current_item = {} lines = content_block.split('\n') for line in lines: if line.startswith('TITLE:'): current_item['title'] = line.replace('TITLE:', '').strip() elif line.startswith('SOURCE:'): current_item['source'] = line.replace('SOURCE:', '').strip() elif line.startswith('URL:'): current_item['url'] = line.replace('URL:', '').strip() elif line.startswith('DESCRIPTION:'): current_item['description'] = line.replace('DESCRIPTION:', '').strip() elif line.startswith('CONTENT:'): current_item['content'] = line.replace('CONTENT:', '').strip() elif 'content' in current_item: # Add to existing content current_item['content'] += ' ' + content_block # Add the last item if current_item and 'title' in current_item: news_items.append(current_item) # Group news items by category ai_tech_items = [] ai_business_items = [] ai_research_items = [] for item in news_items: title = item.get('title', '').lower() description = item.get('description', '').lower() # Simple categorization based on keywords if any(kw in title + description for kw in ['business', 'market', 'company', 'investment', 'startup']): ai_business_items.append(item) elif any(kw in title + description for kw in ['research', 'study', 'paper', 'university']): ai_research_items.append(item) else: ai_tech_items.append(item) # Create sections with subsections sections = [] # AI Technology section if ai_tech_items: tech_subsections = [ Subsection( title=item['title'], source=item['source'], url=item['url'], content=f"{item.get('description', '')} {item.get('content', '')[:500]}..." ) for item in ai_tech_items ] sections.append(Section( name="AI Technology Developments", description="Recent advancements in AI technology and applications", information="Cover the latest developments in AI technology.", subsections=tech_subsections )) # AI Business section if ai_business_items: business_subsections = [ Subsection( title=item['title'], source=item['source'], url=item['url'], content=f"{item.get('description', '')} {item.get('content', '')[:500]}..." ) for item in ai_business_items ] sections.append(Section( name="AI in Business", description="How AI is transforming industries and markets", information="Focus on business applications and market trends in AI.", subsections=business_subsections )) # AI Research section if ai_research_items: research_subsections = [ Subsection( title=item['title'], source=item['source'], url=item['url'], content=f"{item.get('description', '')} {item.get('content', '')[:500]}..." ) for item in ai_research_items ] sections.append(Section( name="AI Research and Studies", description="Latest research findings and academic work in AI", information="Cover recent research papers and studies in AI.", subsections=research_subsections )) # If no items were categorized, create a general section if not sections: general_subsections = [ Subsection( title=item['title'], source=item['source'], url=item['url'], content=f"{item.get('description', '')} {item.get('content', '')[:500]}..." ) for item in news_items ] sections.append(Section( name="Latest AI News", description="Roundup of the latest AI news from around the web", information="Cover a range of AI news topics.", subsections=general_subsections )) return {"sections": sections} except Exception as e: print(f"Error in orchestrator: {str(e)}") # Fallback plan if structured output fails fallback_sections = [ Section( name="Latest AI Developments", description="Overview of recent AI advancements and research", information="Summarize the latest AI developments from the provided content.", subsections=[] ) ] return {"sections": fallback_sections} def llm_call(state: WorkerState): """Worker writes a section of the blog with subsections for each news item""" section = state['section'] # Generate section header without ID for cleaner markdown section_header = f"## {section.name}\n\n{section.description}\n" # If there are subsections, process each one subsections_content = "" llm = get_llm() if section.subsections: for idx, subsection in enumerate(section.subsections): # Generate subsection using LLM subsection_prompt = f""" Write a detailed subsection about this AI news item: Title: {subsection.title} Source: {subsection.source} URL: {subsection.url} Content to summarize and expand on: {subsection.content} Keep your response focused on the news item and make it engaging. Use markdown formatting. """ subsection_content = llm.invoke([ SystemMessage(content="You are writing a subsection for an AI news blog. Write in a professional but engaging style. Include key details and insights. Use markdown formatting."), HumanMessage(content=subsection_prompt) ]) # Format subsection with title and source (without ID tags) formatted_subsection = f"### {subsection.title}\n\n" formatted_subsection += f"*Source: [{subsection.source}]({subsection.url})*\n\n" formatted_subsection += subsection_content.content subsections_content += formatted_subsection + "\n\n" else: # If no subsections, generate the full section content section_content = llm.invoke([ SystemMessage(content="Write a blog section following the provided name, description, and information. Include no preamble. Use markdown formatting."), HumanMessage(content=f"Here is the section name: {section.name}\nDescription: {section.description}\nInformation: {section.information}") ]) subsections_content = section_content.content # Combine section header and subsections complete_section = section_header + subsections_content # Return the completed section return {"completed_sections": [complete_section]} def synthesizer(state: BlogState): """Synthesize full blog from sections with proper formatting and hierarchical TOC""" # List of completed sections completed_sections = state["completed_sections"] # Format completed sections into a full blog post completed_report = "\n\n".join(completed_sections) # Add title, date, and introduction today = datetime.now().strftime("%Y-%m-%d") blog_title = f"# AI News Roundup - {today}" # Generate a brief introduction llm = get_llm() intro = llm.invoke([ SystemMessage(content="Write a brief introduction for an AI news roundup blog post. Keep it under 100 words. Be engaging and professional."), HumanMessage(content=f"Today's date is {today}. Write a brief introduction for an AI news roundup.") ]) # Create hierarchical table of contents table_of_contents = "## Table of Contents\n\n" # Find all section headings (## headings) section_matches = re.findall(r'## ([^\n]+)', completed_report) for i, section_name in enumerate(section_matches, 1): # Add section to TOC with auto-generated link # Create a clean anchor from the section name section_anchor = section_name.lower().replace(' ', '-') table_of_contents += f"{i}. [{section_name}](#{section_anchor})\n" # Find all subsections within this section section_start = completed_report.find(f"## {section_name}") next_section_match = re.search(r'## ', completed_report[section_start+1:]) if next_section_match: section_end = section_start + 1 + next_section_match.start() section_text = completed_report[section_start:section_end] else: section_text = completed_report[section_start:] # Extract subsection headings subsection_matches = re.findall(r'### ([^\n]+)', section_text) for j, subsection_name in enumerate(subsection_matches, 1): # Create a clean anchor from the subsection name subsection_anchor = subsection_name.lower().replace(' ', '-').replace(':', '').replace('?', '').replace('!', '').replace('.', '') # Add subsection to TOC with proper indentation table_of_contents += f" {i}.{j}. [{subsection_name}](#{subsection_anchor})\n" final_report = f"{blog_title}\n\n{intro.content}\n\n{table_of_contents}\n\n---\n\n{completed_report}\n\n---\n\n*This AI News Roundup was automatically generated on {today}.*" return {"final_report": final_report} def assign_workers(state: BlogState): """Assign a worker to each section in the plan""" # Kick off section writing in parallel return [Send("llm_call", {"section": s}) for s in state["sections"]]