Appearance
Platform Roadmap
A single view of the Marchay Platform — what's built, what's next, and where we're headed.
Current State
What Exists Today
| Component | Status | Notes |
|---|---|---|
| Agent Architecture | Phase 1 complete | Session State Management built, Orchestrator pattern validated |
| Integration Architecture | Phase 1 complete | Gateway Foundation + Authentication System built |
| Intelligence Architecture | Phase 1 complete | Unified Data Models built, DMC spike complete |
| Application Architecture | Phase 1 complete | Layout and Navigation built, AI Chat partial |
| Time Tracker | Live | MVP deployed at marchay-time-tracker.netlify.app |
What's Working
- Session State Management with LRU cache, Redis, PostgreSQL
- API Gateway with connector abstraction and pipeline
- Inbound consumer auth (API key + secret) with key rotation
- Outbound vendor auth (OAuth 2.0 JWT Bearer for Salesforce)
- Unified data models (Location, Contact, 7 vertical types)
- Schema versioning and migration system
- SQLite storage with FTS5
- Collapsible dark sidebar with React Router
- AI Chat with intent detection and parameter extraction
- Chat history persistence (IndexedDB)
- Context attachment (Salesforce record search)
- DMC data extraction (95% email, 80% phone coverage)
- MCP server with 5 tools, 3 resources, 2 prompt templates
What's Not Working Yet
- No Research Agents deployed (only Orchestrator pattern)
- No rate limiting or quota tracking
- No error handling and resilience (retries, circuit breakers)
- No real-time inbox (polling only, WebSocket planned)
- No cross-vertical search (single vertical only)
- No production deployment
Platform Vision
A fully integrated data platform for Marchay travel advisors and members. The platform combines:
- AI-powered research — Orchestrator + Research Agents find the best options across dining, experiences, transportation, hotels, flights, and DMCs
- Omnichannel communication — Email, WhatsApp, Teams, RingCentral in one inbox
- CRM integration — Direct access to Salesforce data without leaving the console
- Booking hand-off — Search results transfer to Atlantes for booking
- Travel intelligence — Crawler framework indexes luxury travel knowledge across 8+ verticals
Phases
Phase 1: Foundation (Weeks 1-4)
Goal: Core infrastructure operational
| Week | Agent | Integration | Intelligence | Application |
|---|---|---|---|---|
| 1 | Session State Management | Gateway Foundation | Platform Core (shared types) | UI Framework |
| 2 | Queue Implementation | Authentication System | DMC vertical post-spike | Component Library |
| 3 | Worker Process Lifecycle | Containerization | DMC Data Enrichment | GraphQL Client |
| 4 | Agent Abstraction Layer | CI/CD Pipeline | Dining Vertical (acquisition) | Design System |
Milestones:
☑ Session State Management (Agent)
☑ Gateway Foundation (Integration)
☑ Authentication System (Integration)
☑ Unified Data Models (Intelligence)
☑ Layout and Navigation (Application)
☐ Redis and Postgres provisioned
☐ Queue Implementation (Agent)
☐ DMC vertical post-spike (Intelligence)
☐ Component Library (Application)
Phase 2: Core Features (Weeks 5-8)
Goal: End-to-end flow working
| Week | Agent | Integration | Intelligence | Application |
|---|---|---|---|---|
| 5 | Tool Cascade | Error Handling and Resilience | Dining Schema and Storage | Members List and Detail |
| 6 | LLM Orchestration Loop | Rate Limiting and Quotas | Experiences Vertical | Trips List and Detail |
| 7 | Dining Research Agent | REST API Design | Transportation Vertical | AI Chat Core |
| 8 | Research Agent Framework (verticals) | API Documentation | Hotels Vertical | NLP Pipeline |
Milestones:
☐ Orchestrator decomposing requests and dispatching to agents
☐ Dining Research Agent returning restaurant options
☐ AI Chat showing results with availability and pricing
☐ Members and Trips list pages with filtering
☐ Chat sessions persisting across browser restarts
Phase 3: Verticals and Channels (Weeks 9-12)
Goal: Multi-vertical research and communication channels
| Week | Agent | Integration | Intelligence | Application |
|---|---|---|---|---|
| 9 | Experiences Research Agent | Inbound Webhook Processing | Flights Vertical | Chat History and Context |
| 10 | Transportation Research Agent | Outbound Webhook Delivery | Spa & Wellness Vertical | Omnichannel Inbox |
| 11 | Flights Research Agent | Observability Pipeline | Event Tickets Vertical | Channel Integrations |
| 12 | DMC Research Agent | Security Hardening | Platform Core (cross-vertical) | Dialer Component |
Milestones:
☐ 5+ Research Agents operational
☐ 6+ Intelligence verticals with MCP tools
☐ Email and WhatsApp channels live in inbox
☐ Context attachment linking CRM records to AI searches
☐ Full observability (tracing, metrics, alerts)
Phase 4: Production (Weeks 13-14)
Goal: Production deployment and hardening
| Week | Agent | Integration | Intelligence | Application |
|---|---|---|---|---|
| 13 | Scaling and Reliability | Salesforce Connector Readiness | Cross-Vertical Search | Booking Hand-off |
| 14 | Observability and Cost Telemetry | Production Deployment | Production Deployment | Salesforce UI Bundle Deploy |
Milestones:
☐ All services deployed to production
☐ Autoscaling configured and tested
☐ Failover and recovery tested
☐ Cost telemetry reporting
☐ Cross-vertical search working
☐ Booking hand-off creating Quote__c and Booking__c records
☐ Advisors using the platform
Key Dependencies
External Blockers
| Dependency | Impact | Owner | Status |
|---|---|---|---|
| Salesforce custom fields | Blocks Integration endpoints and Application hand-off | Salesforce admin | Pending |
| Partner API access (Resy, OpenTable, Viator, Amadeus) | Blocks Intelligence verticals | Partnerships | In progress |
| Channel Router (email, WhatsApp, Teams, RingCentral) | Blocks Omnichannel Inbox | Backend team | Not started |
| Orchestrator backend (FastAPI + LangChain) | Blocks AI Chat search execution | Backend team | Spike complete |
Infrastructure Requirements
| Service | Required By | Priority | Status |
|---|---|---|---|
| Redis | Agent, Integration | Critical | In use (Docker Compose) |
| Postgres | Agent | Critical | In use (Docker Compose) |
| Doppler | Integration | Critical | In use |
| Container Registry | All | High | Not provisioned |
| Load Balancer | Integration, Intelligence | High | Not provisioned |
Resource Estimates
| Domain | Projects | Epics | Estimated LOE |
|---|---|---|---|
| Agent | 5 | 14+ | ~8-10 weeks |
| Integration | 5 | 15+ | ~8-10 weeks |
| Intelligence | 10 | 30+ | ~460 hours (~13 weeks) |
| Application | 6 | 18+ | ~8-10 weeks |
Total estimated effort: ~13 weeks with parallel work across domains.
What's After Phase 4
Near-term (Q1 2027)
- Operations Console — Separate UI for business development and member management
- Real-time WebSocket updates — Replace polling in Omnichannel Inbox
- Cross-device chat sync — Server-side session storage for chat history
- Additional verticals — More Intelligence verticals based on advisor demand
Medium-term
- Member-facing app — iOS/Android for members to track trips and communicate
- Advanced analytics — Dashboards for trip performance, advisor efficiency, cost optimization
- Partner portal — Self-service for DMCs, restaurants, experience providers
- AI improvements — Fine-tuned models for better intent detection and recommendations
Long-term
- Predictive analytics — ML models for trip success prediction, churn risk, upsell opportunities
- Automated booking — End-to-end booking without human intervention for simple trips
- Global expansion — Multi-language support, regional verticals, local payment methods
Where to go from here
Last updated: August 14, 2026