Technology

The architecture behind
spatial intelligence

Nexma is built on a schema-driven architecture where a single AI agent serves any infrastructure domain. Here's how the layers work together.

Platform architecture

01
AI Layer

Jax AI Agent

Agentic loop powered by Claude — reads, writes, and reasons over spatial data autonomously. Domain knowledge injected through configuration, not hardcoded tools.

  • Claude Opus / Sonnet / Haiku — auto-selected per task
  • Autonomous spatial reasoning and design execution
  • Domain-agnostic — adapts to any infrastructure vertical
  • 6 solver families: MIP, CP, VRP, graph, simulation, heuristic
02
Data Layer

Spatial Data Engine

Unified data layer that serves as the single source of truth. Schema, spatial data, and configuration coexist in a structured representation. Every view renders from it.

  • Structured spatial data with unified schema
  • Automatic generation from ontology definitions
  • All views (map, ontology, tabs) render from the same source
  • AI-native read/write interface
03
Schema Layer

OntologySchema

The type system that defines a domain. Entities, relationships, constraints, expressions, layer config — load a different schema, get a different application. Zero code changes.

  • 8 fully-typed definition files — JSON-serializable, no `any`
  • Entity types, relationships, constraints, expressions
  • Layer config, view config, toolbar config, constants
  • Schema compiler generates tools, validators, and UI
04
Spatial Layer

Map & Visualization

Mapbox GL JS foundation with satellite imagery, 3D terrain, and live intelligence overlays. Real-time rendering of spatial data from the unified data engine.

  • Mapbox GL JS with custom layers and styling
  • ReactFlow for ontology visualization
  • 100+ live feeds: flights, ships, weather, satellites
  • OpenStreetMap geocoding with multi-language support
05
Optimization Layer

Hybrid Solver Architecture

Six solver families dispatched automatically based on problem structure. Browser-first for speed, server fallback for scale. Self-correcting on constraint violations.

  • Browser: GLPK.js (MIP <500 vars), heuristics, VRP WASM
  • Server: FastAPI on Railway — 5 solver families
  • Automatic dispatch based on problem size and type
  • Self-correction loop on constraint violations
06
Platform Layer

Infrastructure

Next.js 15 monorepo with TypeScript, Turborepo, and pnpm workspaces. Clerk auth, Vercel deployment, and strict Biome linting. Mobile app with offline-first architecture.

  • Next.js 15 + React 19 + TypeScript strict mode
  • Turborepo + pnpm workspaces monorepo
  • Clerk authentication with SSO
  • Biome for formatting and linting

Platform capabilities

Every capability ships with every seat. No feature gates, no module add-ons.

Domain-agnostic

Telecom, water, electric, gas, logistics, construction, defense — one platform, any domain through schema plugins.

70+ languages

Jax speaks your language. Full natural language support with voice I/O in 70+ languages including RTL support.

Mobile-ready

Android field app with offline-first sync, dual camera, voice I/O, and GPS tracking. Designed for construction sites.

Enterprise security

SSO/SAML, audit logs, data isolation, encryption at rest and in transit. SOC 2 and ISO 27001 compliance path.

Live intelligence

100+ real-time data feeds — ADS-B flights, AIS ships, weather, satellites, seismic — unified on a single map.

Open formats

Import and export GeoJSON, Shapefile, KML, CSV, and more. No vendor lock-in — your data stays portable.

See the platform in action

Book a technical demo and see how Nexma's architecture translates to real-world infrastructure design in minutes.