from haystack import Pipeline from haystack.components.builders import PromptBuilder from haystack.components.generators import HuggingFaceAPIGenerator from haystack.utils import Secret from pathlib import Path from haystack.core.errors import PipelineDrawingError # Template for generating detailed module content content_template = """ You are an instructional designer. Create a detailed, educational Markdown module titled "{{name}}" that takes {{duration}} to read. - Use **200 WPM** for general topics or **150 WPM** for technical ones. - Target word count: {{word_count}} ±5%. - Output valid **Markdown only** using the structure below. - Do **not** include code fences, extra commentary, or repeat "### End". ### Structure: # {{name}} **Estimated Reading Time: {{duration}}** ## Introduction - Brief intro to the topic. ## Key Concepts {{formatted_subtopics}} ## Explanation - Detailed explanation using paragraphs, bullets, and examples. ## Examples - Include at least one relevant example. ## Summary - Key takeaways. ### End """ # Create pipeline components builder = PromptBuilder(template=content_template) generator = HuggingFaceAPIGenerator( api_type="serverless_inference_api", api_params={"model": "meta-llama/Llama-3.2-3B-Instruct"}, token=Secret.from_env_var("HF_API_TOKEN"), generation_kwargs={"max_new_tokens": 5000, "stop": [""]}, ) pipeline = Pipeline() pipeline.add_component("prompt_builder", builder) pipeline.add_component("llm", generator) pipeline.connect("prompt_builder", "llm") # Save the pipeline as a YAML file and attempt to draw a diagram with open("./pipelines/generate_module_content.yml", "w") as file: pipeline.dump(file) try: pipeline.draw(path=Path("generate_module_content_pipeline.png")) except PipelineDrawingError: print("⚠️ Mermaid diagram could not be generated (likely offline). Skipping pipeline drawing.") except Exception as e: print(f"⚠️ Unexpected error while drawing pipeline: {e}") # Ensure output directory exists Path("./modules").mkdir(parents=True, exist_ok=True) # Detect technical modules def is_technical_topic(module_name, subtopics): keywords = ["API", "database", "code", "technical", "machine learning", "AI", "infrastructure", "engineering", "system", "security", "algorithm"] combined_text = f"{module_name} {' '.join(subtopics)}".lower() return any(keyword in combined_text for keyword in keywords) # Generate content for all modules in the outline def generate_content_for_outline(outline): content_results = {} for module in outline["modules"]: is_technical = is_technical_topic(module['name'], module['subtopics']) reading_speed = 150 if is_technical else 200 minutes = int(module['duration'].split()[0]) word_count = int(minutes * reading_speed) formatted_subtopics = "\n".join([f"- {sub}" for sub in module["subtopics"]]) print(f"Generating '{module['name']}' at {reading_speed} WPM → {word_count} words") try: result = pipeline.run({ "prompt_builder": { "name": module['name'], "duration": module['duration'], "word_count": word_count, "formatted_subtopics": formatted_subtopics, } }) markdown_output = result["llm"]["replies"][0] content_results[module["name"]] = markdown_output # Save to .md file filename = module["name"].replace(" ", "_").lower() with open(f"./modules/{filename}.md", "w", encoding="utf-8") as f: f.write(markdown_output) except Exception as e: print(f"Error generating content for module '{module['name']}': {e}") content_results[module["name"]] = {"error": str(e)} return content_results # Example outline to test the script updated_outline = { "title": "Customer Service Essentials", "modules": [ { "name": "Introduction to Customer Service", "duration": "10 minutes", "subtopics": [ "Definition of customer service", "Importance of customer service", ], }, { "name": "Handling Customer Complaints", "duration": "20 minutes", "subtopics": [ "Types of customer complaints", "How to handle customer complaints", ], }, ], } # Generate content for the updated outline module_content = generate_content_for_outline(updated_outline) print(module_content)