Mina - Meeting Insights & Action Platform

Overview

Mina is an enterprise-grade SaaS platform designed to transform meetings into actionable moments. It provides real-time transcription with speaker identification, voice activity detection, and AI-powered insights to generate comprehensive meeting summaries and extract actionable tasks. Its core purpose is to enhance productivity and streamline post-meeting workflows, aiming to deliver a cutting-edge platform that significantly improves post-meeting productivity in the growing market for AI-powered business tools.

User Preferences

Preferred communication style: Simple, everyday language.

System Architecture

The application utilizes a layered architecture with Flask as the web framework and Socket.IO for real-time communication, following an application factory pattern. The frontend employs a "Crown+" design system with a dark theme, vanilla JavaScript, and Socket.IO client for a modern and accessible UI/UX.

UI/UX Decisions:

• Crown+ Design System: Glassmorphism effects, smooth animations, consistent design tokens.
• Theming: Dark theme, light mode support, system preference detection.
• Accessibility: WCAG 2.1 AA compliance, screen reader support, keyboard navigation, high contrast/large text modes, ARIA labels/roles.
• Emotional UX: Micro-animations and enhanced hover effects.
• States: Comprehensive styling for empty, error, and loading states.
• Calm Motion: Smooth transitions and animations with prefers-reduced-motion support.

Technical Implementations & Feature Specifications:

• AI Intelligence: Auto-summarization, key points, action items (with assignee, priority, due dates), questions tracking, decisions extraction, sentiment analysis, topic detection, language detection, custom AI prompts, AI model fallback.
• AI Copilot: Chat interface with streaming responses, context awareness (semantic RAG, conversation history, activity summaries), prompt template library, suggested actions, citations, multi-step action chaining, and proactive intelligence.
• Analytics Dashboard: Speaking time distribution, participation balance, sentiment analysis, topic trend, action item completion rate, export, custom widgets.
• Sharing & Integrations: Public sharing (link generation, privacy, expiration), embed, email, Slack, team sharing (role-based).
• Transcript Display: Glassmorphism layout, speaker labels, timestamps, confidence indicators, search, export, copy, inline editing, speaker identification, highlighting, commenting, playback sync, keyboard shortcuts.
• Real-time Audio Processing Pipeline: Client-side VAD, WebSocket streaming, server-side processing, OpenAI Whisper API integration, real-time broadcasting, multi-speaker diarization, multi-language detection, adaptive VAD, real-time audio quality monitoring, confidence scoring, non-blocking concurrent transcription.
• Security & Authentication: JWT-based authentication with RBAC, bcrypt, AES-256 encryption, rate limiting, CSP headers, CSRF protection, input validation, enterprise-grade session management with Redis and dual timeouts.
• Performance: Low Word Error Rate (WER), sub-400ms end-to-end transcription latency, dashboard TTI ≤200ms with cache-first bootstrap pattern.
• Task Extraction: Premium two-stage AI-powered extraction with refinement, metadata enrichment, quality gates, and pattern matching fallback.
• Event Ledger & WebSocket Synchronization: Enhanced EventLedger model, EventSequencer, EventBroadcaster with event emitters, and 4 WebSocket namespaces for real-time updates with workspace isolation.
• IndexedDB Caching + Reconciliation: IndexedDB schema with 5 stores, CacheValidator service with SHA-256 checksums and field-level delta comparison, cache-first bootstrap, and 30-second idle sync with drift detection and auto-reconciliation.
• Unified Sync System: Consistent background synchronization across all dashboard pages via unified-sync-init.js.
• PrefetchController: Intelligent background loading with AbortController, deduplication, queue management, LRU cache eviction, always-Promise pattern.
• Archive Functionality: Meeting archival with metadata tracking, event logging, WebSocket broadcasts, toast notifications with undo, restore capability with audit trail.
• AI-Powered Insight Reminders: Predictive AI reminders using GPT-4o-mini to analyze meeting patterns and tasks, real-time WebSocket delivery, smart fallback, toast notifications with action buttons, analyzes overdue tasks/missing follow-ups/recurring patterns, confidence scoring, workspace isolation.
• CROWN⁴.5 Tasks Page: Enterprise-grade task management with offline-first architecture, event-sequenced updates, sub-200ms first paint. Includes PredictiveEngine, QuietStateManager, Deduper, CognitiveSynchronizer, TemporalRecoveryEngine, and LedgerCompactor.
• CROWN⁴.13 Task Performance Optimizations: Industry-leading task management matching Todoist/Asana/Linear standards with SSR-first architecture, sub-200ms first paint, flicker-free tab switching, neutral visibility control, user action lock, and FCP budget validation.
• Low-Latency Transcription Pipeline: Optimized for <2s transcription delivery targeting Otter.ai parity with StreamingTranscriptionService, direct transcription integration, tiered usage enforcement, server-side authentication binding, real-time latency metrics, and background AI service prewarm.
• Speaker Diarization Status (Dec 2025): Current implementation defaults to single speaker ("Speaker 1") during live recording. Root cause: WebM audio cannot be processed as raw PCM samples by the diarization engine. Recommended fix: Two-tier approach requiring server-side FFmpeg decoding (WebM→PCM) for live labels + managed API (AssemblyAI/AWS Transcribe) for post-session reconciliation. Latency budget: ~60-75ms for decode+VAD keeps under 250ms real-time window. Files: services/speaker_diarization.py, services/multi_speaker_diarization.py.
• CROWN⁴.14 Processing Reliability: Enterprise-grade post-transcription reliability with persistent processing state, stuck job detection, processing status API, real-time WebSocket updates, smart status banner, processing metrics logging, and graceful degradation.
• CROWN⁴.7 Execution Drafts: AI-powered artifact generation from meeting transcripts with AI intent detection, multi-channel support (Email, Slack, Jira, Confluence), Eventlet compatibility, circuit breaker protection, and contact resolution.
• Chrome Extension: Browser extension for seamless meeting audio capture from Teams, Zoom, Google Meet, and Webex with chrome.tabCapture API, JWT-based authentication, PCM16 audio streaming, independent session model, platform detection, and Crown+ design.
  • MVP Architecture: Audio capture, JWT auth, PCM16 audio streaming via Socket.IO /extension namespace, independent user sessions, meeting platform detection via DOM monitoring.
• Premium Founder Dashboard: Enterprise-grade business intelligence across 4 pillars:
  • Revenue & Billing: MRR calculation, plan distribution (Free/Pro/Business), subscription tier counts, Stripe integration.
  • Customer Activity: DAU/WAU/MAU engagement metrics, recent signups, top accounts by meeting count, user growth trends.
  • Content & Growth: Blog post pipeline kanban (Draft → Scheduled → Published), email campaign stats, open/click rates, marketing subscriber count.
  • Operational Health: Real-time system metrics (CPU/Memory/Disk), active meetings, task completion rates, progressive disclosure for detailed metrics.
  • Real-time Updates: WebSocket integration on /admin namespace with 30-second polling and live badge indicator.
• Blog Scheduling System: Database-driven blog management with CRUD admin interface, publish scheduling (auto-publish when date reached), status filtering (draft/scheduled/published), email notification triggers.
• GDPR-Compliant Email Marketing:
  • EmailCampaign and EmailEvent models with SendGrid webhook tracking (opens/clicks/bounces).
  • One-click unsubscribe via unique token, preference center for marketing opt-in/out.
  • Batch email processing with proper CAN-SPAM compliance headers.
  • Marketing consent checkbox in registration flow with timestamp and source tracking.
• Footer Pages: About page with founder story, Security page with trust/encryption/compliance messaging, Pricing page with tier comparison.

System Design Choices:

• Backend: Flask with Flask-SocketIO.
• Database: SQLAlchemy ORM (SQLite for dev, PostgreSQL for prod).
• Session Management: Server-side sessions with triple-layer fallback.
• Real-time Communication: Socket.IO for WebSockets with polling fallback.
• Frontend: Bootstrap dark theme, vanilla JavaScript, Socket.IO client.
• Data Model: Session and Segment models.
• Service Layer: Encapsulated business logic (e.g., TranscriptionService, AI Insights Service, MeetingLifecycleService).
• Production Readiness: Scalability, security, reliability, fault tolerance using Redis for horizontal scaling, distributed room management, session state checkpointing, robust error handling, background task retry systems, and Redis failover.
• Continuous Audio Processing: Overlapping context windows and sliding buffer management.
• Advanced Deduplication: Text stability analysis.
• WebSocket Reliability: Auto-reconnection, heartbeat monitoring, session recovery.
• Monitoring & Observability: Sentry for error tracking, BetterStack for uptime, structured JSON logging, SLO/SLI metrics.
• Backup & Disaster Recovery: Automated, encrypted PostgreSQL backups with multi-tier retention.
• Deployment: CI/CD pipeline (GitHub Actions), Alembic migrations, blue-green deployment.
• Production Readiness (Google SRE Standards): Startup validation with fail-fast on missing config, Kubernetes-compatible health endpoints, environment-aware database initialization, comprehensive production runbook.

External Dependencies

AI/ML Services:

• OpenAI Whisper API
• OpenAI GPT-4o-mini
• OpenAI GPT-4 Turbo
• OpenAI GPT-4.1
• WebRTC MediaRecorder

Database Systems:

• PostgreSQL
• SQLAlchemy
• Redis
• pgvector

Real-time Communication:

• Socket.IO

Security & Authentication:

• Flask-Login
• Flask-Session
• Cryptography
• bcrypt
• PyJWT

Audio Processing Libraries:

• NumPy/SciPy
• WebRTCVAD
• PyDub

Web Framework Dependencies:

• Flask
• Bootstrap
• WhiteNoise
• ProxyFix
• Chart.js

Production Infrastructure:

• Gunicorn
• Eventlet

Other Integrations:

• Stripe (Replit Connector - sandbox for dev, live for production via Publishing pane)
• SendGrid
• Sentry
• BetterStack
• Google Calendar