Automatic Meeting Minutes Generation from Transcription
Meeting minutes—a legally significant document, formatted to corporate standard. It's not just a summary: minutes contain date, participant list, agenda, recorded decisions, votes, and signatures. AI-generated minutes must account for these requirements.
Generation Structure
The process consists of three phases:
Phase 1 — Metadata Extraction: transcription yields date/time, participants (with titles if mentioned), agenda (explicit or recovered from topics).
Phase 2 — Content Structuring: each agenda item → discussion → decision / vote / deferred. LLM processes each section separately for accuracy.
Phase 3 — Template Formatting: result is inserted into corporate template Word (python-docx) or Google Docs (API). Numbering of decisions, date formatting per organization standard.
Templates and Customization
Different organizations have different minutes format requirements. System supports template set with variables:
{{meeting_date}} | {{meeting_number}} | {{chairperson}}
{{agenda_items[]}} | {{decisions[]}} | {{action_items[]}}
Templates are stored as DOCX with bookmarks—python-docx fills bookmarks with extracted data. Final document is sent to participants automatically or queued for approval.
Quality and Control
Auto-generated minutes always go through UI review before finalization: participant clicks "Approve" or makes edits. System learns from edits—corrected fields are logged to improve prompts.