Analysts at a software development company have developed a data processing pipeline to automate the creation of technical specifications using AI agents. The process involves gathering project documentation, meeting notes, and source code from Git repositories to provide context for AI models.
The pipeline utilizes OpenAI's Whisper for transcribing meeting audio into text, followed by an LLM to extract key requirements and decisions. For document conversion, the team uses Pandoc, an open-source utility that transforms various formats like .docx into Markdown, which is better suited for AI analysis.
To generate final documents in the required .docx format, the team employs custom VBA macros in Microsoft Word. These macros automatically apply corporate styles, headings, and formatting to the Markdown output generated by the AI, ensuring compliance with internal standards.
Implementing this workflow has significantly improved efficiency. Processing a one-hour meeting now takes 10–20 minutes instead of 1–2 hours, while document formatting takes 5–10 minutes. The team reports a 30% reduction in requirement errors and a decrease in total time for drafting technical specifications from days to hours.
