Skip to content

PROJECT_STATUS.md

Latest commit

 

History

History
141 lines (116 loc) · 9.42 KB

PROJECT_STATUS.md

File metadata and controls

141 lines (116 loc) · 9.42 KB

Arlo Meeting Assistant — Project Status

Last Updated: 2026-03-13 Version: v1.0 Spec: See /SPEC.md for the authoritative feature specification and version milestones.

Overview

Open-source Zoom Apps starter kit for building intelligent meeting assistants using RTMS (Real-Time Media Streams) — no meeting bot required.

Current state: The v1.0 UI is feature-complete. The frontend has been decomposed from a monolithic component into a multi-view architecture with HashRouter, 14 views, 5 context providers, and a shared AppShell. All Figma Make designs have been ported into the CRA + plain CSS codebase. In-client OAuth PKCE and browser-based web OAuth (Marketplace install) are both working. OS dark mode detection is implemented. New API endpoints support the home dashboard, upcoming meetings, user preferences, and participant event tracking. RTMS auto-start runs at the provider level (MeetingContext) so transcription begins as soon as the user is authenticated and in a meeting, regardless of which view is active. Upcoming meetings are fetched from Zoom's REST API with per-meeting auto-open toggles via the open_apps API. Participant events (join/leave) are tracked with an inline timeline in InMeetingView and a swimlane visualization in MeetingDetailView. Meetings started by RTMS before the user opens Arlo are automatically attributed to the real user via RTMS operator ID. See SPEC.md for the full feature inventory.

Technology Decisions

Component Choice Rationale
Runtime Node.js 20+ Best ecosystem for Zoom SDK and RTMS
Frontend React 18 + CRA 5 Industry standard, Zoom App compatible
UI Library @base-ui/react + plain CSS Unstyled, accessible, CSS custom properties
Routing react-router-dom@6 (HashRouter) Hash-based routing safe for Zoom iframe
Icons lucide-react Lightweight, tree-shakeable icon library
Fonts Source Serif 4 + Inter (self-hosted) Zoom WebView blocks Google CDN fonts
Backend Express.js (JavaScript) Simple, flexible, well-documented
Database PostgreSQL 15+ Full-text search (GIN index), JSON support
ORM Prisma Type-safe, great migrations, modern
AI Provider OpenRouter Free models (Gemini Flash), no API key required
RTMS SDK @zoom/rtms v1.0.2 Class-based Client API, multi-meeting support
Auth Zoom OAuth (PKCE) Native to platform, httpOnly session cookies
Containerization Docker Compose Easy local dev, portable

Open Questions

Question Options Status
Redis required? Redis vs in-memory Testing needed (currently in-memory)
File storage for VTT Local disk vs S3 Starting with local
Vector DB for RAG? pgvector vs FTS only Starting without (FTS works well)
Deployment target Railway, Render, Fly.io Document all options

Code Statistics

  • Backend: ~2,850 lines (9 route files, 4 services, middleware) — JavaScript/Express
  • Frontend: ~11,600 lines (15 views, 6 contexts, 1 hook, 6 UI primitives, 14 shared components, 5 feature verticals) — React 18 + @base-ui/react + plain CSS
  • Feature Verticals: ~6,400 lines across 5 verticals (general, healthcare, legal, sales, support)
  • RTMS: ~370 lines (ingestion worker) — @zoom/rtms v1.0.2
  • Documentation: 15+ guides including reusable Zoom Apps skills
  • Total: ~15,000+ lines of production-quality code

Progress Summary

Completed

  • Project foundation (monorepo, Docker Compose, Prisma schema, env config)
  • Zoom OAuth PKCE flow with encrypted token storage
  • RTMS integration (webhook handlers, transcript ingestion, @zoom/rtms v1.0.2)
  • Live transcript display with WebSocket broadcast (< 1s latency)
  • AI features: summary, action items, chat Q&A (OpenRouter)
  • Auto-suggestions at configurable meeting duration
  • Meeting history, rename, delete
  • Highlights CRUD with timestamps and tags
  • Full-text search (PostgreSQL GIN index)
  • WebVTT export
  • Frontend UI migration to @base-ui/react (Feb 2026)
  • README, ARCHITECTURE.md, TROUBLESHOOTING.md, CLAUDE.md
  • v1.0 multi-view architecture — HashRouter, 14 views, 5 context providers, AppShell with shared header
  • In-client OAuth PKCE flow — useZoomAuth hook, session restoration via GET /api/auth/me, JWT fallback
  • OS dark mode detection with localStorage override and theme toggle
  • Self-hosted fonts — Source Serif 4 + Inter WOFF2 files (Zoom WebView blocks Google CDN)
  • New API endpoints — /api/home/highlights, /api/home/reminders, /api/ai/summary, /api/meetings/:id/export/markdown
  • UI primitives — Button, Card, Badge, Input, Textarea, LoadingSpinner
  • LiveMeetingBanner — "Return to live transcript" sticky banner
  • MeetingCard — Reusable meeting card component with live badge support
  • v1.0 Figma UI implementation (Feb 2026):
    • Search Results View — Full /search route with query highlighting, empty/initial states
    • Delete Meeting Dialog — Confirmation dialog + trash button in meeting detail
    • Rename Meeting — Inline title editing with pencil icon, PATCH API save
    • Transport Controls — 3-state (live/paused/stopped) with red recording dot, orange pause badge, destructive stop button
    • Guest No-Meeting — Redesigned with 3 feature cards and "Connect with Zoom" CTA
    • Guest In-Meeting — Enhanced with live badge, summary skeleton, faded transcript preview, CTA card
    • Home Dashboard — Weekly digest, action items with checkboxes, recurring topic badges, AI-powered weekly digest and cross-meeting insights
    • AI-generated meeting titles — Sparkle icon generates descriptive title from transcript/summary
    • Participant Timeline — 5th tab in MeetingDetail with colored swimlane bars
    • Settings View — Transcription toggles, AI provider/model/API key config, test connection
    • Pause/Resume RTMS — Real pauseRTMS/resumeRTMS SDK calls, rtmsPaused state in MeetingContext
  • Provider-level RTMS auto-start — Auto-start logic moved from InMeetingView to MeetingContext so it fires regardless of which view is active. User lands on HomeView with LiveMeetingBanner linking to the live transcript. Context-aware routing via getRunningContext() guards InMeetingView.
  • Chat notices for transcription lifecycle — Automatic Zoom chat messages for start/pause/resume/stop/restart with per-event toggles and customizable templates
  • Upcoming meetings + auto-open — Zoom REST API integration (GET /v2/users/me/meetings, POST /meetings/{id}/open_apps) with reusable zoomApi service (token auto-refresh, 401 retry). UpcomingMeetingsView with per-meeting auto-open toggles, info/warning banners, sticky bottom bar. Home dashboard shows top 3 upcoming meetings. Settings auto-open toggle reveals compact meeting list. Toast undo support. Requires meeting:read, meeting:write:open_app scopes and ZOOM_APP_ID env var.
  • Web OAuth redirect flow — Browser-based OAuth for Marketplace installs (GET /api/auth/start → Zoom OAuth → GET /api/auth/callback). Three new views: LandingPageView, OnboardingView, OAuthErrorView.
  • Participant event tracking + timeline — ParticipantEvent DB model, swimlane timeline visualization (ParticipantTimeline component), inline events in InMeetingView. Initial roster filtering via firstTranscriptReceived flag.
  • Meeting attribution via RTMS operator ID — Orphaned meetings (created by system user during RTMS auto-start) are automatically reassigned to the real user when they open Arlo.
  • Industry Verticals (Mar 2026) — 5 specialized modes with custom UI components:
    • General (Default): MeetingSummary, KeyMoments, DecisionsLog, OpenQuestions, ParticipantStats, SmartBookmarks
    • Healthcare: SOAPNotesPanel, PatientContextCard, ClinicalAlerts, QuickActions, PreviousSessionsCard, HealthcareTagsSummary
    • Legal: ContradictionDetector, LegalTermsPanel, ExhibitTracker, PrivilegeMarkers
    • Sales: DealTracker, QualificationSignals, CompetitorMentions, CommitmentsPanel
    • Support: SentimentMeter, EscalationAlerts, ResolutionTracker, AgentAssist
    • VerticalContext provider with accent colors, terminology customization, feature flags
    • VerticalSelectorView for first-run setup, Settings toggle for switching

Not Yet Done

  • Automated tests (manual testing only)
  • CI pipeline
  • Demo video
  • Security audit
  • GitHub issue templates
  • End-to-end fresh install test
  • Post-meeting standalone web app (v1.5 goal)
  • Industry Vertical AI Integration — Vertical components currently use demo data. Future work:
    • Real-time AI extraction for medical terms, legal terms, competitor mentions
    • Sentiment analysis from transcript segments
    • Auto-detection of decisions, questions, action items
    • Integration with existing AI service for vertical-specific prompts

Next Actions

  1. End-to-end testing on fresh install
  2. Add basic smoke tests
  3. Add GitHub issue templates
  4. Create demo video
  5. Security audit (OWASP checklist)
  6. Public launch

Related Documentation