QLAB
HARVARD UNIVERSITY INTELLIGENCE PROJECTWHY STRUCTURE
The world is recorded faster than we can analyze it.
Every day, terabytes of new observation are recorded, optical and radar imagery, weather, ships, flights, prices, events.
We automates knowledge creation from this data. We train geospatial foundation models and world-action models to synthesize and run simulations.
FUEL MOISTURE −5% ADDS, AT 72H+193.7 ACRES (10.5%)
At 72 hours a 5% drop in fuel moisture adds 194 acres (10.5%) and 1.7 housing units inside the footprint.
PYRETECHNICS ON LANDFIRE 2.4.0 · POTENTIAL SPREAD ONLY, NO SUPPRESSION · HOURLY CURVES FROM THE RUN'S OWN ARRIVAL-TIME RASTER · FRAMES: SENTINEL-2 / SENTINEL-1 AT 39.70°N 120.82°W, 44 KM AWAY — SAME LANDSCAPE, NOT THE BURN
Research becoming infrastructure.
Structure is backed by Harvard University's QLAB Intelligence Project, NVIDIA Inception, and Google for Startups.

NVIDIA INCEPTION
STARTUP PROGRAMGOOGLE FOR STARTUPS
STARTUP PROGRAMFOUNDATION GEOSPATIAL MODEL
Modeling world dynamics and change.
Our foundation geospatial models fuse imagery, sensors, maps, economics, infrastructure, events, and provenance into fused world state then learns how those states evolve.
OBSERVATIONS → BELIEF STATE → DYNAMICS → POSSIBLE FUTURES
ACTUAL DATA · ALPHAEARTH 2025 GLOBAL EMBEDDING FIELD · 64-D · NATURAL-COLOUR ADAPTER FIT TO NASA BLUE MARBLE · 10,681,024 ADAPTER-VALID EO OBSERVATIONS

Simulate alternative pasts and futures.
Run simulation of different interventions in specific times and locations. Highlighted overlays show year-over-year changes given specified situations.




CASE STUDIES
DEPLOYED RESEARCH // DECISION SYSTEMS
Case Studies
We use automated analysis and foundation geospatial models into operational systems for institutions making mission critical decisions.

Alternative Histories for Model Evaluation
We generate alternative histories to evaluate models against more than the single path the world happened to take.
- MISSION
- MODEL VALIDATION
- DATA
- IBTrACS V04R01 + ERA5
- RESOLUTION
- 0.25° × HOURLY
- OUTPUT
- LOSS-DISTRIBUTION ENSEMBLE

Automated Analysis of Cascading Effects
Granular automated data analysis applies the equivalent of a coordinated team of PhD researchers to each portfolio question.
- MISSION
- PORTFOLIO DATA ANALYSIS
- DATA
- GDELT 2.1 + CLIENT TABLES
- SERVICES
- 70+ CONNECTED SOURCES
- OUTPUT
- REPRODUCIBLE EVIDENCE PACK

Cleaning and Standardization of International Data Sources
Our geospatial foundation models augment, clean, and infer missing data across international markets where US-grade records do not exist.
- MISSION
- GLOBAL EXPOSURE COMPLETION
- DATA
- ALPHAEARTH + SENTINEL-1/2
- EMBEDDING
- 64-D GEOSPATIAL FIELD
- OUTPUT
- AUGMENTED RISK ATTRIBUTES

Simulation as a Service
We provide simulation as a service to the Army for mission analysis, readiness planning, and evaluation across controlled scenarios.
- MISSION
- Redacted
- OUTCOME
- Redacted
- DETAILS
- Redacted

Operation and Scenario Planning
We help special operations teams run scenarios, predict likely outcomes, and plan across operational constraints before deployment.
- MISSION
- Redacted
- OUTCOME
- Redacted
- DETAILS
- Redacted

Analyst Decision Support
We make analysts more effective by measuring and understanding possible outcomes before they happen.
- MISSION
- Redacted
- OUTCOME
- Redacted
- DETAILS
- Redacted

Foundation World and Geospatial Models
We train foundation world models and geospatial models for forecasting, simulation, and empirical research.
- MISSION
- SPATIOTEMPORAL FORECASTING
- DATA
- 10,681,024 EO OBSERVATIONS
- WINDOW
- 2022–2025 ANNUAL FIELDS
- OUTPUT
- FORECAST-READY WORLD STATE

Research Support and Partnerships
We partner with faculty at leading institutions to publish emerging research in simulation and spatial intelligence.
- MISSION
- JOINT METHODS RESEARCH
- CORPUS
- 243,968 PRETRAIN LOCATIONS
- MODE
- CO-DEVELOP + PUBLISH
- OUTPUT
- REPRODUCIBLE PAPER + CODE

Faster Experiments and Conclusions
We provide data-analysis support so researchers can design experiments, test hypotheses, and reach defensible conclusions faster.
- MISSION
- EXPERIMENT ACCELERATION
- DATA
- 7,732,347 WEEKLY ROWS
- CONTEXT
- 266,432,936 RICH-CONTEXT ROWS
- OUTPUT
- TESTED RESULT + METHODS LOG
AGENT DATA SCIENTIST TEAM
Our data science team of agents.
Design the experiments, find and clean data, run analyses, do robustness checks, and visualize the reuslts. They run and integrate directly with your other tools.
AGENT ROSTER — 8 PROCESSES, 1 PIPELINE
- RESEARCH LEADquestion · hypotheses · success criteria
- DATA ENGINEERfind · join · clean · validate
- GEOSPATIAL SCIENTISTlocation · scale · imagery
- MODELING SCIENTISTmodels · forecasts · simulation
- STATISTICIANbaselines · uncertainty · calibration
- CAUSAL SCIENTISTidentification · causal designs
- REVIEWERleakage · alternatives · limitations
- COMMUNICATORfinding · methods · evidence
01 ▸ 08 · EACH AGENT HANDS ITS RESULT TO THE NEXT
THE HUMAN + INSTITUTIONAL STATE · NIGHT LIGHTS FROM ORBIT · NASA
WORK WITH US
Simulate alternative worlds with and cause and effect understanding.
Our models and data scientist team turns observations and one-off analysis into knowledge for decision support.
HELLO@STRUCTURE.MLCDAO · TRADEWINDS SOLUTIONS MARKETPLACE
Awardable
and Ready for mission.
Our foundation geospatial models and analyst services are Awardable on the Department of War Chief Digital and Artificial Intelligence Office Tradewinds Solutions Marketplace, providing government customers a streamlined path to evaluate and acquire our technology.






