TriNetra View Demo

When Infrastructure Fails, People Are Waiting. Know Who Needs You First.

TriNetra combines satellite imagery analysis, infrastructure dependency mapping, and real-time cascade modeling to help emergency teams make faster decisions in the critical first 72 hours.

The Three Eyes

Damage classified in under 90 seconds.

Cascade calculated in under 400ms.

Restoration ranked before teams mobilize.

The problem

When Maria made landfall, 95% of Puerto Rico lost power.

Hospitals had 72 hours of generator fuel. Utility crews had no map of what depended on what.

TriNetra was built so the next response team doesn’t fly blind.

0.0M
people affected
0
months to full power restoration
$0B
in damages

TriNetra: The Three Eyes

Three layers of intelligence. One decision surface.

Eye 1: Damage Detection

See What Broke, The Moment Imagery Arrives

The pipeline pulls pre- and post-event RGB stacks from Microsoft Planetary Computer (NAIP with Sentinel-2 L2A fallback), crops per asset, and runs a six-channel EfficientNet-B4 classifier in PyTorch. CUDA accelerates inference when a GPU is present; otherwise the model runs on CPU.

Eye 2: The Dependency Graph

One Substation Goes Down. Find Out What Goes With It.

Behind every city is an invisible web: power flows to water treatment, which flows to hospitals, which flow to the people who need them most. TriNetra maps 14 types of infrastructure dependency so when something breaks, you instantly see everything it takes with it.

Eye 3: Priority Scoring

Not a Map Full of Red Dots. A List That Tells You What to Do.

After a damage observation runs through the cascade engine, its priority score scales the log of total downstream population impacted by the strictest affected criticality tier, model confidence, damage severity, and an urgency term tied to how soon the first downstream asset hits time-to-failure, so restoration order reflects real failure horizons, not dot density.

01
Sinbanpo Substation Powers Gangnam hospital cluster, 6 metro stations
94.7
02
Jamsil Water Treatment Plant Sole potable supply for 312,000 residents south of the Han
88.2
03
Dongjakgu Emergency Relay Tower Backhaul for 14 cell sites across three flooded districts
79.5

Cascade engine

priority score

Integrations

Built on the Data You Already Depend On

HIFLD-backed facility registries, Voronoi service areas inferred in code, PostGIS edge generation, NetworkX cascades, and GeoJSON-first APIs you can query today.

Infrastructure data
HIFLD ARCGIS REST TIGER CENSUS API FEMA NSS
Live ArcGIS queries for substations, hospitals, towers, plants, shelters; TIGER block groups plus decennial counts; Puerto Rico demo generator when you need synthetic loads.
Spatial analysis
POSTGIS VORONOI SHAPELY SRID 4326
Clipped Voronoi diagrams for substations and water plants, persisted as polygons; power and water edges from ST_Within containment, comms from ST_DWithin tower proximity.
Graph & cascade
NETWORKX BFS PYDANTIC
Directed DiGraph rebuilt from Postgres; breadth-first downstream walk with failover minutes, depth caps, and typed CascadeAnalysis payloads.
API & visualization
FASTAPI GEOJSON MAPBOX GL
Async asset CRUD, dependency create and list plus GeoJSON edges, bbox filters, radius area loads, stored cascade analysis; Mapbox GL JS with terrain, extrusions, data-driven styling.

Deployment

Built for Government. Ready for Deployment.

Geospatial database
POSTGRES 16 POSTGIS 3.4 GEOALCHEMY2 ALEMBIC
postgis/postgis:16-3.4 in Docker Compose; GeoAlchemy2 geometry columns; Alembic-managed schema including cascade persistence.
Async API tier
FASTAPI ASYNC SQLA PYDANTIC V2
Async SQLAlchemy sessions, typed request and response models, OpenAPI from routers, health checks, and paginated GeoJSON list endpoints.
Container & access
DOCKER COMPOSE X-ADMIN-KEY CORS
One-file database bring-up for local and demo stacks; graph rebuild and spatial reinference gated by X-Admin-Key when configured.

The Next Disaster Won’t Wait. Neither Should Your Response.