Electron-Based Meeting Notes Assistant

Overview

Built a second Electron application focused on online meeting note-taking assistance. The system prioritized cost efficiency through local ML models, resilient recording pipelines, and sync between desktop and web clients — spanning Google Meet, Microsoft Teams, Slack Calls, and Zoom.

Tech stack

• Electron
• TypeScript
• Node.js
• Whisper
• SQLite
• Next.js
• NestJS
• MongoDB
• Supabase
• Stripe
• AWS S3
• New Relic
• Mouseflow
• Google Calendar
• FastAPI

Features implemented

• Local Whisper model for cost-efficient transcription instead of external LLM services
• Local audio diarization for speaker separation
• Error-handling and retry mechanisms for failed recordings
• Local queue system for concurrent recordings using Node.js workers
• SQLite for local storage with periodic server sync
• High-performance Next.js website with SEO and social sharing (OG images, SSR, dynamic routing)
• Supabase authentication with sync between web and desktop users
• User activity tracking via New Relic and Mouseflow
• MongoDB full-text and fuzzy search for global search
• Meeting sharing for authenticated and unauthenticated users
• S3 presigned URLs for secure file uploads
• Customizable AI agent behavior for meeting analytics (summaries, highlights, and more)
• Google Calendar sync with auto-launch one minute before meetings
• Custom RMS-based silence detection achieving 95%+ accuracy (replacing Silero VAD)
• Stripe subscription management
• Data archiving for inactive users
• Support for Google Meet, Microsoft Teams, Slack Calls, and Zoom
• Migration of a large Python FastAPI backend to NestJS
• Release notes for each app version

Key achievements / impact

• Optimized infrastructure cost by moving transcription to a local Whisper model
• Improved silence detection accuracy to 95%+ with a custom RMS-based approach
• Collaborated with a remote team across UI, API integrations, and backend systems
• Successfully migrated a large FastAPI backend to NestJS