Methodology

How we collect, process, and classify defense R&D award data.

Data Source

All data is sourced from SBIR.gov, the official U.S. government repository for Small Business Innovation Research and Small Business Technology Transfer awards.

17,265
Total Awards
2020-2024
Date Range
DoD
Agency Filter

Dual-Dimension Classification

Each award is classified along two independent dimensions. This enables analysis by technology type (what is being developed), platform domain (where it will be applied), or both dimensions together.

Dimension 1
Technology Type

The underlying technology or capability being developed.

18 categories
Dimension 2
Platform Domain

The application environment or system type.

8 categories

Technology Categories (18)

Category Description
Sensors & ImagingRadar, lidar, cameras, infrared, multispectral, acoustic sensors
Computing & SoftwareProcessors, embedded systems, algorithms, mission software
Propulsion & EnginesJet engines, rockets, electric propulsion, turbines
Artificial IntelligenceMachine learning, neural networks, computer vision, autonomy
Materials & StructuresComposites, armor, coatings, structural materials
Power & Energy StorageBatteries, fuel cells, power generation, energy harvesting
CommunicationsRadio, networks, data links, antennas, spectrum
CybersecurityNetwork defense, encryption, threat detection
Manufacturing & ProductionAdditive manufacturing, machining, quality control
Biotechnology & MedicalMedical devices, biosensors, human performance
Modeling & SimulationDigital twins, training systems, physics simulation
Navigation & PNTGPS, inertial navigation, positioning, timing
Thermal & EnvironmentalHeat management, cooling, thermal protection
Guidance & ControlFlight control, targeting, stabilization, seekers
Directed EnergyLasers, high-power microwave, particle beams
Electronic WarfareJamming, SIGINT, electronic attack/protection
Warheads & EnergeticsExplosives, munitions, warhead design
Command & ControlBattle management, decision support, C2 systems

Platform Categories (8)

Category Description
AerospaceFixed-wing aircraft, rotorcraft, UAVs, eVTOL
SpaceSatellites, launch vehicles, orbital systems
Ground VehiclesTanks, trucks, unmanned ground vehicles
Naval SurfaceShips, boats, surface vessels
UnderseaSubmarines, UUVs, torpedoes
Missiles & MunitionsGuided missiles, bombs, projectiles
Soldier/DismountedIndividual equipment, wearables, infantry systems
Enterprise/InfrastructureNetworks, data centers, logistics, facilities

Classification Method

Awards are classified using Claude (Anthropic's large language model) via the Claude API. Each award title is analyzed and assigned to one or more categories in each dimension based on semantic understanding of the content.

Model
Claude 3 Haiku
Success Rate
99.8% (17,233 of 17,265)

Why AI classification?

  • Semantic understanding captures context beyond keyword matching
  • Consistent classification criteria across all 17,265 awards
  • Ability to assign multiple categories when appropriate
  • Handles domain-specific terminology and acronyms

Limitations

1. Title-based classification. Analysis is based on award titles only. Full abstracts are not consistently available in the source data.
2. AI classification accuracy. While the model achieves high consistency, some edge cases may be misclassified. Estimated accuracy is 90-95%.
3. Category boundaries. Some technologies span multiple categories. The classification reflects the primary focus of each award.
4. SBIR scope. SBIR awards represent early-stage R&D investments and do not capture total DoD technology spending.
5. Classified programs. Awards for classified programs may have generic or obscured titles, affecting classification accuracy.

Data Updates

Data is manually refreshed from SBIR.gov on a periodic basis. The current dataset includes awards through December 2024. Classification is re-run when new data is added to ensure consistency across the full dataset.

Data: SBIR.gov | AI Classification: Claude (Anthropic) | Coverage: 2020-2024