📅 2025-06-10 — Session: Developed Editorial Workflow Automation
🕒 21:20–23:15
🏷️ Labels: Editorial Workflow, Automation, AI, Azure Ml, Jinja, Openai
📂 Project: Dev
⭐ Priority: MEDIUM
Session Goal
The session aimed to design and implement a structured editorial workflow pipeline for processing articles from raw input to publication, leveraging automation tools and AI-driven solutions.
Key Activities
- Structured Editorial Workflow Pipeline: Outlined a framework for processing articles, detailing each stage’s inputs, processes, and outputs.
- Systematic Prompt Definitions: Developed prompt definitions for various article processing stages, including CSV parsing and agenda generation.
- Function Calling in OpenAI API: Explored function calling for structured data handling in OpenAI API.
- Fuzzy Row Selection in Azure ML: Implemented a fuzzy row selection mechanism using OpenAI’s function calling within Azure ML.
- Handling Function Calling Limitations: Addressed limitations and strategies for function calling in LLMs.
- Function Schema for Article Parsing: Designed a robust schema for parsing and clustering news articles.
- Jinja Prompts for Article Processing: Developed multiple Jinja prompts for tasks like seed card generation, draft article creation, and metadata annotation.
- Pipeline Design for LLM Screening: Outlined a minimal starter pipeline for LLM screening.
- AzureML PromptFlow Pipeline Design: Streamlined a pipeline for parsing and clustering articles in AzureML.
Achievements
- Established a comprehensive editorial workflow with automation and AI components.
- Developed templates and schemas to enhance data processing and article generation.
- Enhanced AzureML pipelines with improved data parsing and mapping techniques.
Pending Tasks
- Implement the designed pipelines and test their integration with existing systems.
- Further refine prompt definitions and schemas based on testing outcomes.