On-Device AI & Machine Learning

Large language models, vision analysis, voice control, and translation — running entirely on your Android device with zero cloud dependency.

On-Device Large Language Models

Tactical AI integrates MediaPipe for running state-of-the-art LLMs directly on Android hardware. Models are downloaded separately and run with full offline capability.

Model Parameters RAM Required Best For
Google Gemma 3N 1B / 4B 4GB / 6GB General reasoning, spatial queries
Microsoft Phi-4 Mini 3.8B 6GB Code generation, analysis
MiniCPM 2B 4GB Resource-constrained devices

🎯 Zero Cloud Dependency

All AI processing happens on-device. No data leaves your device, making this suitable for classified networks, denied environments, and operations where cloud connectivity is unavailable or prohibited.

AI-Enabled Spatial Queries

Combine the power of DuckDB Spatial with natural language. Ask questions about your geospatial data in plain English and receive structured SQL queries and results — no SQL expertise required. The on-device LLM translates intent to spatial operations.

Example AI Spatial Queries

Vision Language Service

TensorFlow Lite scaffolding provides image analysis capabilities for field intelligence. Analyze geotagged photos, drone imagery, and surveillance feeds using on-device computer vision models for object detection and scene classification.

Capability Status Offline Use Case
Object Detection Available Vehicle, structure, person detection
Scene Classification Available Terrain type, land use analysis
Image Captioning Available Automatic photo description
OCR Text Extraction Available Sign, document reading

Voice Control & Actions

Speech recognition enables hands-free operation of ATAK map functions. Issue voice commands to navigate, drop markers, query data, send messages, and trigger map actions — critical for operators with gloves or in high-tempo environments.

🔨 Field Operations Use Case

During dismounted operations with gloves, use voice commands to drop markers, request route calculations, and send status updates without touching the screen. Text-to-speech reads incoming C2 messages aloud for eyes-free situational awareness.

Text-to-Speech & Translation

On-device speech synthesis reads messages and alerts aloud. Real-time language translation enables cross-language communication in multinational operations, all processed locally without network access.

Supported Translation Languages

Google Gemma 3N Microsoft Phi-4 MiniCPM MediaPipe TensorFlow Lite Speech-to-Text Text-to-Speech
← Geospatial Analytics Next: Real-Time Data →