Overview
Evergreen is the comprehensive business platform I invented and built from the ground up at Midtown Home Improvements. I prototyped it in two weeks as a basic CRM, and over the last 4+ years it has evolved into a full-scale enterprise ecosystem—650K+ lines of code across 8+ interconnected services—that powers every aspect of the business across all 5 branches.
The ecosystem includes the core Nuxt application, a real-time call center platform (Twilio + Convex) used daily by telemarketers, Nest.js microservice APIs, a custom ML prediction system (Oracle), a mobile field app with 200+ daily users (Cypress), a learning management system, and background job infrastructure processing tens of thousands of jobs daily.
The Problem
When I joined Midtown, the company was struggling with:
- Fragmented systems: Different tools for sales, scheduling, customer management, and billing
- Data silos: No single source of truth for customer information
- Manual processes: Staff spending hours on repetitive tasks that could be automated
- Limited visibility: Leadership had no real-time insight into business performance
The Solution
Rather than cobbling together off-the-shelf solutions, I designed and built Evergreen—a unified platform that serves as the single source of truth for the entire organization.
Core Modules
CRM & Contact Management
The heart of Evergreen, managing the entire customer lifecycle:
- Lead Aggregator: Integrated with external lead sources for automatic ingestion
- Contact Management: Complete customer history in one place
- Oracle ML Scoring: Custom model predicts sale probability and optimal rep assignment
- Property Enrichment: Automatic enrichment with geo, property, and market data
Call Center Platform
Real-time communication hub used daily by the telemarketing team:
- Browser-based Softphone: Twilio VoIP calling directly in browser
- Convex Real-time State: 9 tables for agent status, call queues, SMS threads
- Two-way SMS: Real-time inbox with thread locking and DNC compliance
- AI Call Grading: Gemini-powered analysis of call recordings
- Inbound/Outbound: Handle incoming calls and structured outbound campaigns
Business Intelligence Dashboard
Real-time analytics giving leadership instant visibility:
- Live KPIs: Sales metrics, conversion rates, revenue tracking
- Department Scorecards: Branch, production, service, telemarketing, canvass
- Trend Analysis: Historical comparisons and forecasting
- CMO Dashboard: Separate executive analytics service
Client Portal
Self-service access for customers:
- Production Job Tracking: Real-time project status updates
- Document Center: File uploads, contracts, and photos
- Service Tickets: Customer support request management
- Communication: Direct messaging with project team
Oracle - Custom ML Sales Prediction
A custom-trained machine learning model that predicts sale outcomes:
- Sales Prediction: Predicts likelihood of closing based on lead characteristics
- Rep Matching: Ranks which sales rep has the best chance of closing each lead
- Lift Calculation: Measures improvement over baseline random assignment
- Data Sources: Trained on historical appointments, geolocation, soft credit, and market data
Cypress - Mobile Field App
Production mobile app (iOS/Android) with 200+ daily active users:
- Real-time GPS Tracking: Admin dashboard shows 200+ canvassers live on map
- Door Knock Logging: Every knock tracked with geo coordinates
- Field-to-Call-Center Pipeline: Leads submitted from field go directly to TM queue
- Scaling Challenges Solved: Batching, optimization for high-frequency location updates
Additional Modules
- AI/Automagic: Call audio grading, transcription, grammar suggestions via Gemini/OpenAI
- TaskMaster: Background job engine (Nitro.js/BullMQ) processing tens of thousands of jobs daily
- LMS: Learning management system for employee onboarding and training
- SMS Platform: Real-time two-way SMS with Convex-powered inbox and DNC compliance
- Helpdesk: Internal ticket system with Linear integration
- Production Tracking: End-to-end job lifecycle with installer portal
Architecture Decisions
Why a Custom Solution?
Building from scratch allowed us to:
- Perfect fit: Every feature designed for our specific workflows
- Deep integration: Seamless data flow between all modules
- Rapid iteration: Deploy improvements weekly, not quarterly
- Cost control: No per-seat licensing or feature limitations
Technical Stack Choices
Nuxt 4 + Nest.js Microservices: Nuxt on the frontend for excellent DX and SSR capabilities. Nest.js powers separate microservice APIs for specific domains.
PostgreSQL + Read Replicas: Rock-solid relational database with 77 models, 200+ indexes, and read replicas for scaling query-heavy operations.
Convex + Ably: Real-time state management for the call center with 9 Convex tables handling agent status, call queues, and SMS threads.
Redis + BullMQ: Powers TaskMaster, processing tens of thousands of background jobs daily with dynamic module loading.
Python + scikit-learn: Oracle ML system for sales prediction, trained on proprietary company data.
Capacitor: Cypress mobile app delivering iOS/Android experience for 200+ daily field users.
Prisma ORM: Type-safe database access with excellent migration tooling.
Architecture Patterns
The ecosystem has evolved based on real-world needs:
- Hybrid Real-time: Convex for hot ephemeral state, Postgres for permanent storage
- Monolithic Core: Main Nuxt application is a well-structured monolith for simplicity
- Microservices: TaskMaster, Oracle, LMS, and CMO Dashboard as separate services
- Event-driven: Domain-organized event handlers for complex workflows
- Queue-based Processing: BullMQ with dynamic module loading for extensibility
Impact
Since launching Evergreen:
- 40% reduction in lead response time
- 25% increase in conversion rates
- 60% less time spent on administrative tasks
- Real-time visibility for leadership decision-making
Lessons Learned
Building Evergreen taught me invaluable lessons about enterprise software:
- Start with workflows, not features: Understanding how people actually work is more important than building what they say they want.
- Invest in foundations: The time spent on a solid data model and clean architecture pays dividends every day.
- Iterate with users: Weekly releases with real user feedback beats quarterly big-bang deployments.
- Documentation is a feature: Good docs reduce support burden and improve adoption.
What's Next
Evergreen continues to evolve:
- Expanding Oracle ML with additional data sources and prediction models
- Enhanced AI-powered call quality scoring and coaching
- Deeper real-time analytics with natural language queries
- Continued scaling optimizations for growing user base
