Switchain Services

Appearance

Refactoring Switchain: From Legacy to Microservices ​

1. WHY (Why) - The Problem ​

The legacy system after 7 years ​

Technical challenges:

  • Monolithic architecture in Meteor.js difficult to maintain
  • Accumulated technical debt compromising development
  • Limited scalability for individual components

Business challenges:

  • Unsustainable operational costs ($40-50K/year on rates alone)
  • Operational risks in a system processing billions of USD
  • Slowed development velocity
  • Difficulty onboarding new developers

The reality: A critical system that works, but can't evolve at the pace the business needs.

2. WHAT (What) - The Solution ​

Complete refactoring with modern architecture ​

From this:

Meteor.js Monolith
├── admin/
├── interface/
├── crons/
├── partners/
└── lambdas/ (💸 $50K/year)

To this:

Microservices + DDD
├── apps/
│   ├── admin (Next.js)
│   ├── sw-api
│   ├── rates (💰 $85/month)
│   └── specialized workers
└── packages/
    ├── shifts (bounded context)
    ├── wallets (bounded context)
    ├── rates (bounded context)
    └── more contexts...

Modern stack:

  • Node.js/TypeScript with Bun
  • Domain-Driven Design
  • Hexagonal Architecture (Ports & Adapters)
  • Monorepo organized by domains

3. THE STRATEGY: Mikado Method + Value-First ​

What is the Mikado Method? ​

Inspired by the Japanese game Mikado (pick-up sticks):

  • 🎋 You have a pile of intertwined sticks
  • 🎯 Goal: Remove one stick without moving any other
  • ⚠️ If something moves, you lose your turn
  • 🔄 Keep trying until all sticks are removed

Applied to software:

Legacy System (pile of sticks)
      ↓
Extract one responsibility (one stick)
      ↓
Without breaking anything else ← CRITICAL!
      ↓
Everything else keeps working
      ↓
Repeat until migration complete

Key principle: Never break the system in production. The reality: It's like tying your shoelaces… while running.

Our adaptation: Value-First Approach ​

We don't migrate by technical ease, but by business impact.

Prioritization based on: ​

  1. Economic value (cost savings, revenue generation)
  2. Risk reduction (critical components)
  3. Enablers (unlocks other migrations)
  4. Learning (strategy validation)

🚀 Success Case #1: Rates System ​

Context:

  • January 2025 Hackathon as catalyst
  • Need to renew contract with Vercel
  • Rates system running on expensive lambdas

The problem:

💸 Current cost: $40,000 - $50,000 per year
📈 Trend: Growing with volume
⚠️ Risk: Unsustainable long-term

The action:

  • First component extracted from legacy
  • Migrated to optimized dedicated server
  • Implemented with new architecture

The result:

✅ New cost: $85 per month = $1,020 per year
💰 Savings: ~$49,000 annually
📊 ROI: 98% cost reduction
⚡ Bonus: Better performance and control

Strategy validation ​

What we learned:

✅ It works: Mikado Method is effective in production ✅ Immediate ROI: No need to wait until the end to see value ✅ Confidence: Team validated that migration is viable ✅ Momentum: Quick wins generate energy to continue

📊 The Real Result: Zero Downtime Migration ​

The ultimate proof:

📅 Period: January 2025 → August 2025 (8 months)
💰 Daily volume: ~$2,000,000 USD
📈 Total volume: ~$480,000,000 USD migrated
⏱️ Downtime: 0 minutes

The technical feat:

Migrating a complete system that processes $2M daily, without stopping for a single second, is like tying your shoelaces while running a marathon.

Migrated components:

  • ✅ Rates System → $49K/year savings
  • ✅ Wallets API → New architecture
  • ✅ Specialized workers → Better scalability
  • ✅ Core business logic → DDD implemented
  • ✅ System virtually complete in production

Next wave of migrations:

  • Admin panel (improved user experience)
  • Final optimizations
  • Complete legacy deprecation

4. WHO (Who) ​

Project stakeholders ​

Critical system impacting:

  • 👥 End users: Crypto exchanges without interruptions
  • 🤝 Partners: Integrated APIs and services
  • 🛠️ Tech team: Better developer experience
  • 📊 Operations: Reduced costs, greater efficiency
  • 💼 Business: Agility for new features and markets

Key requirement: Team training in DDD and hexagonal architecture

5. WHEN (When) ​

Project timeline ​

Q4 2024 ✅ Completed

  • Analysis and design of new architecture
  • Definition of bounded contexts
  • POC and technical validation

Q1 2025 ✅ Completed

  • January Hackathon: Rates System migrated
  • Implementation of core services
  • Pattern establishment
  • Zero-downtime migration begins

Q2 2025 ✅ Completed

  • Migration of main APIs
  • Wallets and Shifts contexts
  • Processing $2M daily in parallel

Q3 2025 ✅ Virtually completed

  • Workers and auxiliary services
  • Performance optimization
  • System migrated in August
  • Zero downtime throughout the process

Q4 2025 🎯 In progress

  • Final refinement
  • Legacy system deprecation
  • Documentation and handover

6. WHERE (Where) ​

Scope and context ​

Industry: Cryptocurrency exchange platform

System scale:

  • 💰 Billions USD processed
  • ⏱️ 7 years of continuous operation
  • 🌍 Global users and partners
  • 🔒 24/7 critical system

Technical scope:

From: 6 monolithic applications
To:   15+ specialized microservices
      + 8 bounded contexts
      + Scalable architecture

7. HOW (How) ​

Migration principles ​

Zero Downtime

  • Deployments without interruptions
  • Fast rollback if necessary
  • Feature flags for granular control

Backwards Compatibility

  • Compatible APIs during transition
  • Coexistence period legacy + new
  • Gradual migration of consumers

Data Integrity

  • Continuous data validation
  • Synchronization during transition
  • Exhaustive edge case testing

Risk Mitigation

  • Real-time monitoring
  • Proactive alerts
  • Contingency plan for each phase

Technical stack ​

Runtime & Language:

  • Node.js with Bun (speed and efficiency)
  • TypeScript (type safety)

Frontend:

  • Next.js 14 + Shadcn/UI
  • React Server Components

Architecture:

  • Domain-Driven Design
  • Hexagonal Architecture (Ports & Adapters)
  • Independent Bounded Contexts

Monorepo:

  • Bun workspaces
  • Shared packages for cross-cutting concerns
  • Independent apps per service

Success metrics ​

Technical:

  • 📊 Code coverage: 80%+
  • ⚡ Build time: -50%
  • 📦 Bundle optimization per app
  • 🧹 Technical debt reduced

Business:

  • ✅ Uptime: 99.9%+
  • 🚀 Response time: +30% improvement
  • 📈 Development velocity: +40%
  • 🐛 Bug rate: -60%
  • 💰 Costs: We already saw the rates case

Continuous validation:

  • Legacy vs new comparison in parallel
  • A/B testing of critical features
  • User feedback loops

Real Risks and Challenges ​

Technical risks faced ​

⚠️ Migration complexity

  • Critical system in production 24/7
  • Mitigation: Small phases, fast rollback
  • Result: Zero downtime achieved ✅

⚠️ Loss of functionality

  • Risk of omitting legacy features
  • Mitigation: Feature parity checks, exhaustive testing
  • Result: All functionalities migrated ✅

⚠️ Performance regression

  • Possible temporary degradation
  • Mitigation: Continuous monitoring, benchmarks
  • Result: Performance improved ✅

😰 The biggest human challenge: The Support team ​

The problem:

January - August: Coexistence of TWO systems
├── Legacy Admin (old functionalities)
└── New Admin (new functionalities)

The daily reality of support:

  • 🔄 "Do I do this in the old or new admin?"
  • 🤔 Memorizing where each functionality is
  • 📝 Duplicate documentation
  • 🐛 Debugging in two parallel systems
  • 😵 Constant confusion for the team

Impact on developers:

  • Maintaining two codebases
  • Bugs appearing "in which system?"
  • Duplicate testing
  • Constant context switching

🎉 The light at the end of the tunnel ​

September 2025:

✅ Admin migration complete
✅ Single source of truth
✅ Support can be happy again
✅ Developers can sleep peacefully

Special thanks:

To the Support team for their infinite patience during 8 months of "which admin do I use?"

Now there's only ONE admin. 🙏

Current status:

  • Single system in production
  • Incremental improvements ongoing
  • Pending details that don't affect operations
  • Team morale: 📈📈📈

Conclusion ​

Speed matters: A temporal analysis ​

Original system (Legacy):

📅 Intensive development: 3-4 years
👥 Team building from scratch
🛠️ 2018 technologies
📚 Learning on the go

New system (Complete refactoring):

📅 Total migration: 8 months (January - September 2025)
💰 Value from month 1: $49K saved/year
👥 Team with accumulated know-how
🤖 AI as key accelerator

The equation:

3-4 years → 8 months = 5x faster

🤖 The AI factor: The invisible copilot ​

How AI accelerated us:

January 2025:

  • DDD boilerplate generation
  • Architecture suggestions
  • Legacy code refactoring

September 2025:

  • More powerful and accurate AI
  • Better context understanding
  • More effective assisted debugging

Continuous evolution:

AI in January ───→ AI in September ───→ AI today
     │                   │                  │
   Good            Very good          Excellent

Reflection:

If we started today, with current AI and acquired knowledge, we would do it faster and better.

Lesson: Development speed is not linear when you combine:

  • Accumulated know-how
  • Well-thought architecture
  • Exponentially evolving tools

The journey virtually completed ​

What we've achieved:

  • ✅ Modern and scalable architecture designed and implemented
  • ✅ Migration strategy validated in production
  • ✅ $49K saved/year on rates alone
  • ✅ $480M processed during migration
  • ✅ Zero downtime in 8 months of active migration
  • ✅ System migrated in September 2025
  • ✅ 5x faster than building from scratch
  • ✅ Team trained, confident… and happy
  • ✅ Support with a single admin (finally!)

What's next:

  • 🎯 Complete legacy system deprecation
  • 🎯 Final performance optimizations
  • 🎯 Documentation and knowledge transfer
  • 🎯 Prepare platform to scale 10x
  • 🎯 Apply learnings to future projects

The three technical feats ​

1. The availability feat:

Migrating a critical system that processes $2M daily, for 8 months, with zero downtime, is like changing a plane's engine… mid-flight.

2. The speed feat:

Recreating in 8 months what originally took 3-4 years, significantly improving it, is like running a marathon at sprint speed.

3. The human feat:

Maintaining two systems, two admins, two realities, without losing sanity or service, is like juggling… with chainsaws.

The future has arrived ​

Expected impact:

A system that has moved billions and lasted 7 years, now with modern architecture prepared for:

  • The next 7+ years of growth
  • Scale 10x without structural changes
  • Easily adopt new technologies
  • Onboard new developers in days, not months

The promise fulfilled:

  • ✅ Faster system
  • ✅ Costs drastically reduced
  • ✅ More productive team
  • ✅ Platform ready for the future
  • ✅ And all without stopping for a second

Q&A ​

Questions?

Contact: [Your information]

Documentation: [Link to repo/docs]