Introducing The

Marchay Platform

A fully integrated data platform for Marchay travel advisors and members.

Unified Experience

AI-Powered Research

Proprietary Intelligence

Platform Vision

One platform. Every advisor. Every member. Every trip.

Core Concepts

Unified Consoles & Apps

One place for research, communication, booking, etc.

AI-Native Research

Intelligent agents that find, filter, personalize, and recommend.

Data Advantage

Proprietary intelligence that competitors can't replicate.

Extensible by Design

Modular architecture for new verticals, channels, and partners.

Current State

Future State

Fragmented data

Single source of truth

Point solutions

Unified platform

Manual workflows

Intelligent automation

Internal only

Member-facing experiences

Discovery Results

19 hours. 4 working prototypes. Validation across every major platform capability.

Spike

What It Accomplished

Value to the Platform

Research Agents

Built a proof of concept dining research agent capable of searching availability across multiple user-definable data sources and using custom tools.

Validates the AI orchestration pattern for research, extendable from dining to any domain advisors need to query.

Integrations Interface

Built a pluggable API gateway that acts as a backbone for SF and accepts adapters for external integrations. An adapter covers 11 CRM CRUD functions.

Proves a single integration architecture scales to every external system with no rebuilds, just new adapters.

Travel Intelligence

Built foundational intelligence facilities and scrapers for DMCs. Collected 231 real DMC profiles with contact info, creating a searchable intelligence layer.

Establishes the data foundation for every vertical in the platform, proving we can acquire and expose domain data at scale.

Advisor Console

Built a working advisor UI on Salesforce with functional AI chat, inbox scaffolding, and access to various CRM data including members and trips.

Confirms the advisor experience can live where advisors already work, eliminating adoption friction.

View Demo

What's Next

These POCs are not throwaway prototypes, rather the foundation for the actual platform. Each will be hardened and expanded, not replaced.

Platform Architecture

Five layers, each independently extensible.

UI Layer

Front-end application.

AI Layer

Agents classify requests, deploy agents, and aggregate responses.

Integration Layer

Enables connection external platforms.

Data Layer

External APIs and real-time scraping.

Intelligence Layer

Proprietary datasets, built by crawlers, exposed via MCP.

Advisor Console

React + Vite + shadcn/ui · Salesforce UI Bundle

AI Chat · Unified Inbox · Member/Trip Data · Hand-off

search & chat

inbox & hand-off

Orchestration

FastAPI + LangChain

Orchestrator Agent · Vertical Router

Integrations

WhatsApp · Teams · RingCentral · Email · Atlantes

Webhook receivers · Message normalization · WebSocket push

fan-out

Agentic Researchers

Dining · Experiences · Transportation · DMC

Vertical-specific Agents

via MCP

Travel Intelligence

via MCP Server

Crawler framework · Data Warehousing · MCP

query

Data Sources

Partner APIs + Scraping

Resy · Yelp · Viator · GetYourGuide · Uber · Blacklane · Booking.com

Key Decisions

Confirmed choices and their rationale.

Architecture

StackPython + Node.jsPython for agents, Node for API/integrations

Data ProtocolMCPPurpose-built for AI agent consumption

Agentic FrameworkLangChainProven in Spike 1, parallel fan-out

AI

Model ProviderOllama (Dev) Fireworks.ai (Prod)Hosted services, no GPU infra needed. Dev tooling provides parity with zero API costs

Model - OrchestrationLlama 3.2Cheap, follows the rules

Model - ResearchLlama 4 Maverick + ScoutExcels in research and synthesis

Infrastructure

CloudDigital OceanSimple, cost-effective for one engineer

DatabasePostgreSQLSingle store for messages, crawled data, CRM

Code HostingGitLabEnterprise grade Git with ecosystem of tools

CI/CDGitLab CIIntegrated with code hosting

ObservabilityPrometheus + GrafanaReasonable long-term operating cost

Roadmap Approach

Rolling planning process — reviewed quarterly.

Now

Until EOY 2026

Commited execution on tasks that can be planned now.

Defined strategic themes, refined from previous learnings.

Later

Q2+ 2027

Strategic, but higher uncertainty. Subject to more validation.

Quarterly Review Cadence

→ Review progress and outcomes
→ Reassess priorities
→ Move initiatives from Next → Now when clearly definable

Quarterly Release Cadence

→ Quarter is decomposed to 3 four-week sprints
→ One large release per quarter
→ Smaller monthly improvement releases post-launch

Now: Build the MVP

Four sprints resulting, one release with feedback cycle. 12 weeks to production.

Sprint 1

Platform Foundation + Core Data

Weeks 1–4

CI/CD, secrets management, monitoring, API gateway, containerization. First research agents live.

Sprint 2

Integrations + Console

Weeks 5–8

All verticals and channels functional. Unified inbox operational. All communication adapters deployed.

Sprint 3

Hardening + Pilot

Weeks 9–12

Production deploy, staging soak, security hardening, advisors onboarded for UAT.

Sprint 4

Adoption + Refinement

Weeks 13–16

Advisor feedback, monthly improvement release, v1.1 prioritization.

Focus shifts from active development to continuous improvement.

Core Deliverables

→Production-ready platform

→3+ research verticals

→4 communication channels

→End-to-end hand-off

→UAT advisors onboarded

Next: Expand Capabilities

Q1 2027 focus areas, refined from Phase 1 learnings.

Platform Expansion

Q1 2027 Focus Areas

→ Managing booking fully inside advisor console*
→ Additional platform integrations (Hotels, Flights, new verticals)
→ Expanded workflows beyond research and hand-off
→ Advisor-facing capabilities and UX improvements
→ Operations console (UI and automations for BD, member management)

*Dependent on Atlantes API stability and maturity

Refined From Phase 1

What We'll Learn

→ Which verticals drive the most value
→ Which integrations are most critical
→ Where advisors need more support
→ What data is most frequently accessed

Takeaway

Built on Phase 1 foundation, refined by real advisor workflows.

Later: Evolve Platform

Q2+ 2027 and beyond — strategic potential.

Strategic Capabilities

→
Member App
iOS and Android for members to view trips, communicate with advisors, access itineraries
→
Self-Serve Booking
Bookable experiences for end users, designed for private label spin-off
→
AI-Enabled Experiences
Intelligent recommendations, personalized suggestions, predictive concierge
→
Advanced Analytics
Advisor performance, member engagement, operational insights
→
Ecosystem Expansion
New verticals, new partners, new geographies

Private Label Opportunity

White-Label Spin-Off

→ Self-serve booking can be spun off as a white-label product
→ Target private banks, luxury brands, exclusive membership organizations
→ Enabled by the data and integration foundation built in earlier phases

Takeaway

Each phase unlocks the next — platform value compounds over time.

Operating Cost Estimates

Projected monthly costs for 25 users, 8x5 usage.

Infrastructure (Digital Ocean)

Component	Choice	Mo.
Servers	4x Droplet 4GB	$96
Database	Managed PostgreSQL	$30
Queue	Managed Redis	$15

DevOps

Component	Choice	Mo.
CI/CD	GitLab CI (Free)	$0
Observability	Prometheus + Grafana	$0

* LLM cost depends on query volume. 25 advisors × ~100 queries/day × ~1K tokens avg.
** WhatsApp: Free for service messages, $0.005–0.14 per template message depending on country.
*** RingCentral: Assumes advisors are using at least Ring EX plan.

AI & Data

Component	Choice	Mo.
LLM Provider (Production)	Fireworks.ai	$50–150^*
LLM Provider (Dev)	Ollama	$0
Agent Framework	LangChain	$0

Integrations

Component	Choice	Mo.
WhatsApp	Meta Direct API	$0^**
Email / Teams	Microsoft Graph API	$0
Phone / SMS	RingCentral	$0^***
Salesforce	UI Bundle + GraphQL	$0

Estimated total monthly: $191–291/mo

Success Metrics, Risks & Dependencies

What success looks like and what could get in the way.

Success at End of Phase 1

→ Modular platform foundation established
→ Functional integrations between communication platforms + Atlantes
→ Data flows reliably between all systems and stores
→ Advisor console ready to roll out to advisors
→ Clear path into future phases

Key Risks

Risk	Likelihood	Impact	Mitigation
Partner API approval delays	Very High	High	Self-serve APIs can be live week 1. Partner approvals take 1–4 weeks. Agents degrade gracefully.
One engineer, no buffer	High	High	Agentic tooling provides 3–4x leverage. Scope is adjustable — fewer verticals or channels if needed.
Atlantes hand-off issues	Medium	Very High	Engage Atlantes team in week 1, ensure they can deliver within phase 1 window.
WhatsApp verification delay	High	Medium	Apply day 1, long lead time. If not verified by week 6, ship with Email, Teams, RingCentral.

Leadership Alignment & Next Steps

The path is clear. We need alignment to move forward.

What We Need From Leadership

→ Confirm roadmap direction
→ Align on Phase 1 priorities
→ Approve operating costs
→ Establish review cadence (short and long term)
→ Direction on platform naming

Immediate Next Steps

→ Submit partner API applications (week 1)
→ Engage Atlantes team on next steps
→ Identify UAT group and advisor lead
→ Adopt roadmap and project management model

We've reached the end.

View Platform Docs