Evidence

Engineering

Three kinds of thing sit below, and it is worth being clear which is which: libraries anyone can install, commercial products we sell and run, and internal systems we are opening up over time. What connects them is not a shared runtime — they do not compose into a platform — but an engineering position held consistently: Elixir and the BEAM for concurrency, Rust NIFs for computational geometry, PostGIS and H3 for spatial storage and indexing, and MapLibre for rendering, used across projects rather than swapped between them.

Some of this is published as open source, some is a commercial product, and some is internal work we are opening up gradually. Each entry below says which, so nothing here reads as a promise it is not.

One engineering position, not a shared runtime

These systems do not compose into a platform. What repeats across them is the choice of tools — which is the claim this page is making, and the tags below are how you check it.

  • Elixir8/13
  • Rust NIF3/13
  • PostGIS3/13
  • MapLibre GL JS3/13

Geospatial systems

The only cross-project integration among these four is GeofenceX calling RoutingEX for map matching. Each of the others is standalone, loaded independently rather than composed into a shared platform.

  • Quorix

    Flagship

    A self-hosted, multi-tenant Location Intelligence platform and an open-stack alternative to CARTO. It lets a user upload geospatial data, publish styled vector-tile layers, compose maps with analytical widgets, run PostGIS-backed spatial analysis, and embed the result elsewhere. Built on Elixir/Phoenix with Rust NIFs (via Rustler) for the geometry-heavy work, a partitioned and row-level-secured PostGIS features table, single-query MVT tile generation, and a React + MapLibre GL + deck.gl front end with SAML 2.0 SSO. It has real CI and a substantial test suite. It is the most mature project in the group.

    • Elixir/Phoenix
    • Rust NIF
    • PostGIS
    • MapLibre GL
    • deck.gl
    • React

  • TileserveX

    A production vector tile server in Elixir with a Rust NIF for MVT encoding, serving PostGIS, MBTiles, and PMTiles sources through a layered cache. Its real differentiator is standards breadth: it implements OGC API Tiles Part 1, OGC API Features Parts 1 through 3 (including CQL2 spatial predicates), and OGC API Processes Part 1, where comparable open-source tile servers typically stop at tiles alone.

    • Elixir
    • Rust NIF
    • PostGIS
    • MBTiles
    • PMTiles
    • OGC API
    Architecture and benchmarks

  • RoutingEX

    A production self-hosted OSM routing engine covering route, matrix, isochrone, TSP, map matching, and GTFS transit, with a Rust NIF implementing contraction hierarchies for query-time performance.

    • Elixir
    • Rust NIF
    • OpenStreetMap
    • GTFS

  • GeofenceX

    Real-time geofencing built on H3 hexagonal indexing instead of per-polygon containment checks against a growing dataset, PostGIS with TimescaleDB hypertables for GPS history, and Phoenix Channels pushing zone-entry and zone-exit events live.

    • Elixir
    • H3
    • PostGIS
    • TimescaleDB
    • Phoenix Channels

Enterprise systems

Operations and asset management built around a specific industry's model rather than configured out of a generic platform. Both are commercial products we sell and operate; neither is open source, and neither is planned to be.

  • SANA Asset Management

    Commercial product

    An operations and asset management suite for adult day health and senior care. It models the participant as the entity the operation revolves around, with caregivers, agencies, services, transportation, case managers and medication schedules attached to it, so the reporting a waiver programme requires comes out of the operational data instead of being reassembled from exports each month.

    • Enterprise software
    • Operational reporting

  • OpenEAM

    Commercial product

    An enterprise asset management system for utilities, in active development. Asset registries, work orders and maintenance planning for organisations whose assets are distributed across a service territory rather than sitting in one building.

    • Enterprise software
    • Asset registry

Applied machine learning

Forecasting systems where the prediction is a calibrated distribution rather than a single answer, and where being wrong is measurable within hours.

  • Sports Predictor

    Flagship

    An internal forecasting system for football and baseball. Football combines a Dixon-Coles bivariate Poisson baseline, a gradient-boosted Poisson model and a sequence Transformer, with ensemble weights tuned by Optuna against Brier score; baseball uses a multiplicative log-linear model for expected runs plus an XGBoost classifier and a walk-forward starting-pitcher adjustment. Both feed a Monte Carlo engine that runs 50,000 simulations per match to produce the full outcome distribution rather than a scoreline. A Rust service handles ingestion, FastAPI serves the predictions, and a Phoenix LiveView dashboard drives it. Acceptance is calibration — Brier, log-loss, ranked probability score and expected calibration error — never accuracy: any new model has to beat the classical baseline on calibration before it ships.

    Worth knowing
    • Blends model lambdas before simulation rather than averaging their output probabilities, which preserves the exact-scoreline distribution instead of flattening it.
    • An earlier gradient-boosted regression for baseball runs learned to predict the league average; the log-linear reformulation that replaced it is what distinguishes a strong offence from a weak one.
    • Storage, experiment tracking, ONNX export and Rust serving are designed but not built. What runs today is the Python pipeline, the simulation engine, the API and the dashboard.
    • Python
    • FastAPI
    • XGBoost
    • PyTorch
    • Rust
    • Elixir/Phoenix
    • Monte Carlo

Open-source libraries

Released packages, not demos. Each one exists because the mapping layer of a real project needed it.

  • HoloMap

    Open source · Hex

    Declarative MapLibre GL JS map components for the Hologram Elixir framework. A map is a component tree: sources, layers, markers, and controls are components, their appearance is component state, and interaction comes back as Hologram actions — with no JavaScript in the application.

    Install as holo_map

    • Elixir
    • Hologram
    • MapLibre GL JS
    View package

  • MaplibreX

    Open source · Hex

    MapLibre GL JS as declarative Phoenix LiveView components. Maps are built from assigns and handle_event/3 rather than hand-written JavaScript. It has the best test-to-code ratio of our libraries.

    Install as maplibrex

    • Elixir
    • Phoenix LiveView
    • MapLibre GL JS
    View package

  • MaplibreX Svelte

    Open source · npm

    Component-oriented MapLibre GL JS bindings for Svelte 5, written in TypeScript. You describe the map as a component tree and the library keeps that tree and the imperative MapLibre instance in sync — adding, updating in place, and tearing down in the right order.

    Install as maplibrex-svelte

    • Svelte 5
    • TypeScript
    • MapLibre GL JS
    View package

Platform

Our content-platform work has gone through three generations, each answering a different question.

  • ApexCMS

    An API-first design on the Ash framework with an OpenAPI-described public API. It answered how content should feed external clients.

    • Elixir
    • Ash
    • OpenAPI

  • CatalystCMS

    A LiveView-first design with a runtime field API over JSONB. It answered how to model content types at runtime.

    • Elixir
    • Phoenix LiveView
    • JSONB

  • WordexCMS

    Flagship

    The current generation: a self-hosted CMS whose differentiator is an installable component system — self-contained units that register their own routes, API endpoints, channels, and database migrations, with migration collision detection and permission scoping. Components hot-load without a restart, with install and uninstall as true inverses of each other. It carries a large test suite concentrated on the dangerous paths of installing and removing code at runtime, and it has been through a systematic security-hardening pass.

    • Elixir
    • Phoenix
Talk to our team