Enterprise Case Studies: Upgrading Actuarial & Geospatial Platforms to glmap

Overview

🌱 ELI5 (Real-World Impact in Production Apps)
When an insurance or environmental team builds a big Shiny dashboard with leaflet, they quickly hit a wall:

• They have to cap policies on the map at 5,000 or 10,000 dots (max_markers) because drawing all 200,000 policies crashes the browser.
• They have to load 4 different helper packages (leaflet, leafem, leaflet.extras, leafpop) and write custom JavaScript workarounds just to make polygons and popups work.

With glmap, one package replaces all four. Underwriters and actuaries can see every single policy, claim, flood zone, building footprint, and rainfall grid at once, draw a box around any neighborhood, and get instant Total Insured Value (TIV) exposure!

⚙️ Technical Mode (Enterprise Architecture Transformation)
In production R + Shiny deployments (Geonesis catastrophe exposure and Terrascope S3 asset lifecycle management), Leaflet.js forces developers to either cap Snowflake queries (LIMIT max_markers + markerClusterOptions()) or write custom session$registerDataObj() + geojson-vt workarounds. Replacing the rendering tier with glmap eliminates 4 package dependencies (leaflet, leafem, leaflet.extras, leafpop), cuts R worker CPU blocking by 15x–40x, and connects directly to Snowflake / DuckDB Spatial via gl_bbox_to_wkt() and gl_popup_table().

---

Case Study 1: Catastrophe & Underwriting Exposure Platform (Geonesis)

Before (leaflet + leafem + leaflet.extras + leafpop)

In the legacy leaflet architecture, the application had to: 1. Import 4 separate mapping packages and manually concatenate ?key=... onto CARTO tile URLs. 2. Cap policy (ALL_RISKS_GEO) and claim (ALL_CLAIMS_GEO) queries with max_markers and group them into DOM marker clusters (markerClusterOptions()), hiding true spatial TIV concentration until max zoom. 3. Convert terra rainfall rasters to custom RGB hex vectors (grDevices::col2rgb) on the R CPU. 4. Serialize entire flood_zones.geojson and perimeters.geojson layers over the Shiny WebSocket on toggle.

After (glmap GPU Edition)

1. Single-Call API Key Setup:

   library(glmap)
   gl_set_api_key("carto", Sys.getenv("CARTO_API_KEY"))

2. Full-Portfolio GPU Rendering + C++17 Hexagonal TIV Accumulation: Instead of hiding policies behind DOM marker clusters, glmap renders both a C++17 hexagonal TIV accumulation surface (add_hexbin(weight = ~TOTAL_INSURED_VALUE)) and the individual instanced GPU policy points (add_points()) at 60 FPS:

   glmap_proxy("geonesis") |>
     clear_group(c("Risks", "Claims", "TIV Hexbins")) |>
     add_hexbin(
       risks_df, lon = "LONGITUDE", lat = "LATITUDE",
       weight = ~TOTAL_INSURED_VALUE, hex_radius = 0.04,
       palette = "plasma", group = "TIV Hexbins"
     ) |>
     add_points(
       risks_df, lon = "LONGITUDE", lat = "LATITUDE",
       color = ~TOTAL_INSURED_VALUE, palette = "viridis",
       radius = 3.0, group = "Risks"
     ) |>
     add_points(
       claims_df, lon = "LONGITUDE", lat = "LATITUDE",
       color = "#f97316", radius = 5.0,
       popup = gl_popup_table(claims_df, zcol = 1:6),
       group = "Claims"
     )

3. Direct Box-Brush to Snowflake / DuckDB Spatial WKT (gl_bbox_to_wkt()): When an underwriter drags a selection box using glmap’s built-in Box Brush tool, input$geonesis_selected emits both the brushed feature indices and a ready-to-bind EPSG:4326 POLYGON ((...)) WKT string (gl_bbox_to_wkt(input$geonesis_selected)) for Snowflake ST_WITHIN(GEOGRAPHY, TO_GEOGRAPHY(?)) or DuckDB ST_Intersects(geom, ST_GeomFromText(?)).

---

Case Study 2: S3 Geospatial Asset & Release Workstation (Terrascope)

Before (Custom geojson-vt + session$registerDataObj() Workaround)

Because leaflet could not render large .geojson, .fgdb, or .tif release artifacts in Terrascope, the viewer previously required custom JavaScript injection (geojson-vt.min.js) and manual session$registerDataObj() gzip endpoints—and still could not dynamically re-color or hillshade Float32 GeoTIFF hazard rasters in the browser.

After (glmap Universal Asset Previewer)

With glmap, any S3 asset (.tif, .parquet, .fgdb, .geojson) is previewed on the GPU in a single function:

terrascope_preview_asset_gpu <- function(asset_path, color_col = NULL, palette = "viridis") {
  ext <- tolower(tools::file_ext(asset_path))
  m <- glmap(basemap = gl_providers$CartoDB.DarkMatter)

  if (ext %in% c("tif", "tiff", "geotiff")) {
    # Streams raw Float32 raster to WebGL2 with 0ms live palette & hillshade controls
    m |> add_raster(terra::rast(asset_path), palette = "terrain", hillshade = TRUE)
  } else if (ext == "parquet") {
    # Pushes viewport bounding box directly into DuckDB / GeoParquet
    m |> add_duckdb(source = asset_path, color = color_col %||% "#00d4ff", palette = palette)
  } else {
    # Triangulates polygons in C++17 (mapbox/earcut.h) and packs GLMP binary buffer
    v <- sf::st_read(asset_path, quiet = TRUE)
    m |> add_sf(v, fill_color = color_col %||% "#38bdf8", popup = gl_popup_table(v))
  }
}