How carlkho-minerva/omi-hackathon_25mnrv works
Main Components
• Client Interaction Points:
• Omi Device that captures conversation audio and transcripts.
• End-user’s mobile phone or browser accessing the reflection UI.
• Backend API Services:
• Omi Webhook Collector (Cloud Run) that receives memory creation events.
• Daily Reflection Processor (Cloud Function) that aggregates and processes raw data.
• Data Storage:
• Firestore databases storing raw conversation data (‘raw_memories’) and processed reflection data (‘daily_reflections’).
• External Services:
• OpenAI API for AI-powered analysis of conversation data.
• Cloud Scheduler that triggers the processing job daily.
• (In future iterations) Push Notification Service for sending notifications.
File Structure
• The file tree divides the codebase into two main folders—Archive and omi-wrapped.
• The “Archive” folder contains experiments or earlier versions (e.g., memory-testapi, omi-to-notion) that might still illustrate useful patterns like Dockerfiles and Python scripts.
• The “omi-wrapped” folder encapsulates the production-ready application with directories for:
◦ daily-reflection-processor (the Cloud Function).
◦ omi-webhook-collector (the Cloud Run API service with its Dockerfile).
◦ omi-wrapped-frontend (static HTML/CSS/JavaScript that renders the user interface).
◦ docs (detailed documentation including the Firestore schema).
◦ personal-docu (additional documentation or notes).
• Notable configuration files like requirements.txt and Dockerfiles show language and runtime dependencies.
Extract Insights from the README
• Key GCP components (Cloud Run; Cloud Functions; Firestore; Cloud Scheduler; Cloud Storage) are highlighted.
• Integration points such as the Omi Memory Creation Webhook and the OpenAI API for processing are explicitly described.
Final Advice
• Don’t shy away from breaking down the project into as many individual components as possible. The goal is to capture every significant part of the architecture.