Build an Air-Gapped Private Meeting Minutes Pipeline with Docker, FunASR & Ollama

I am deploying an end-to-end air-gapped meeting transcription and minutes pipeline on-premise using Docker Compose with FunASR and Ollama (Qwen2.5).
Generate:
1) [Production `docker-compose.yml`]: Running FunASR (port 10095) and Ollama (port 11434) with local model caching, internal networking, and hotword directory mounts;
2) [End-to-End Automation Script (`meeting_pipeline.py`)]: Takes a meeting audio file, calls local FunASR for transcription + speaker diarization, and streams the transcript to local Ollama to output markdown meeting minutes;
3) [Executive Minutes Extraction Prompt]: Strict schema extracting 1) core consensus and ratified decisions, 2) controversies and open blockers, and 3) an Action Items Markdown table (task, owner, definition of done, due date) without hallucination.