LandScope

The Problem

Mitch Rawlyk had built LandScope as a proof of concept running on Google Earth Pro. Customers wanted it in a browser. The MVP worked but wasn't scalable.

The Approach

Instead of thinking "frontend," "backend," "data," "DevOps" as separate concerns, I thought about The System. What does this thing need to do, end to end, and how do we make it fast and reliable from day one?

Architecture

• Frontend: React + Tailwind + TypeScript + CesiumJS (3D globe)
• Admin Panel: Svelte, gRPC/connect-rpc
• Backend: GoLang + gRPC, no frameworks
• Auth: OpenPolicyAgent, Zitadel v3
• Observability: OpenTelemetry
• Data: Postgres (not PostGIS — business logic on app server)
• Storage: S3-compatible object store
• Processing: All binary, no Python. Direct cgo and pipe integration with open source geospatial tools.
• Full deployment automation and observability out-of-the-box.

Key Decisions

• Binary processing pipeline instead of Python: speed and operational simplicity
• Postgres without PostGIS: keep GIS logic in application layer for flexibility
• OpenTelemetry from day one: production observability isn't optional
• No frameworks: Google-style simplicity, direct control

Results

• Under 20 minutes to process 2,000 acres (most layers in under 3 minutes)
• Production-grade from launch
• Platform handles real customers, real payments, real data
• Architecture scales from 1 to infinity