RecapFlow
Reference Architecture
A reference architecture for automating post-call content generation using n8n, Fathom, and LLM prompt chains — from transcript capture to published deliverables.
12 chapters · Reference documentation · n8n + Fathom + LLM pipeline
Documentation
12 chapters covering philosophy, infrastructure, workflows, operations, and synthesis.
Introduction
What RecapFlow is, who it serves, and how this reference architecture is structured.
Design Philosophy
Core principles guiding the architecture: simplicity, observability, and deterministic automation.
Architecture Overview
End-to-end data flow from call recording to published content — the Capture > Merge > Distill > Publish pipeline.
Infrastructure & Deployment
n8n self-hosting, Docker Compose configuration, and deployment environment setup.
Workstation-Side File Sync
Syncing generated content from the server to the local workstation for review and publishing.
Credentials & Security
API key management, secrets storage, and security considerations for the pipeline.
Workflow: Fathom Transcript Poller
Automated polling workflow that fetches new Fathom transcripts on a schedule.
Workflow: Fathom Manual Lookup
On-demand workflow for manually looking up and processing specific Fathom calls.
Workflow: Merged Call Summarizer
Merges multiple call transcripts and generates consolidated summaries via LLM prompt chains.
Testing & Debugging
Strategies for testing workflows, debugging failures, and validating prompt outputs.
Prompt Engineering
Prompt design patterns, template structure, and iteration strategies for LLM-driven content generation.
Lessons Learned & Future Directions
Key takeaways, trade-offs encountered, and planned improvements for the RecapFlow pipeline.