We designed, developed, and hardened a comprehensive suite of on‑device tactical intelligence tools purpose‑built for modern defense operations. Every component runs entirely on edge hardware — no cloud connectivity, no external API calls, no reliance on remote servers — ensuring full functionality in disconnected, contested, and electromagnetic‑denied environments. The suite integrates counter‑drone detection, natural‑language command and control, naval maritime coordination, and deep entity intelligence into a unified, air‑gapped ecosystem that can be deployed on ruggedized tablets, single‑board computers, and militarized Android devices.
1. VAJRA — Counter‑UAS Multi‑Sensor Detection & Classification
Mission: Detect, classify, and track unauthorized drones (UAS) in real time using fused sensor data, running entirely on embedded edge hardware.
Core Technology:
VAJRA ingests live feeds from multiple sensor modalities — pan‑tilt‑zoom (PTZ) cameras, thermal imagers, and software‑defined radio (SDR) sniffers that capture drone communication signatures (e.g., DJI OcuSync, FPV video links). All processing happens locally on an NVIDIA Jetson Orin module. A custom‑trained YOLOv8n (nano) convolutional neural network runs at 30+ FPS on the visual‑thermal stream, achieving millisecond‑level detection latency. The model was trained on a proprietary dataset of over 150,000 labelled drone images across various weather conditions, backgrounds, and flight postures, with aggressive data augmentation to ensure robustness. A separate 1D CNN processes real‑time spectrogram slices from the SDR, classifying drone RF protocols even if the drone is visually obscured.
Multi‑Sensor Fusion & Alerting:
A lightweight Kalman‑filter‑based tracker fuses bounding‑box detections from YOLOv8n, thermal blobs, and RF‑derived azimuth angles. The fusion engine assigns a probabilistic threat score based on flight behaviour (loitering, high‑speed approach), proximity to no‑fly zones, and RF match confidence. Alerts are generated locally — audio, visual, and haptic — and can be dispatched over tactical networks (MANET radios) to distributed operator nodes. The entire pipeline runs in a Docker container on the Jetson with less than 15W power draw, allowing battery‑operated surveillance posts to run for days. VAJRA operates fully disconnected; no cloud processing is required for model inference, sensor fusion, or threat scoring.
2. KAVACH — SLM360 Natural Language Command & Control Platform
Mission: Enable commanders and operators to issue complex tactical commands and query the battlefield picture using natural speech, without typing or menu diving, all processed offline on a local device.
Core Technology:
KAVACH embeds SLM360, a bespoke 360‑million‑parameter small language model optimized for military terminology and constrained grammar. The model was distilled from a larger foundation model, then quantized to INT8 and fine‑tuned on a curated corpus of operational orders, NATO reporting standards, and Indian Army procedure manuals. It runs entirely on‑device (laptop, tablet, or single‑board computer) using ONNX Runtime with CPU acceleration, requiring no GPU and no external API.
Capabilities:
An operator speaks a command like “Assign Alpha Team to grid 32Q ND 1573 6284, suppress enemy position at the treeline, call sign Vikram, fire mission HE, 3 rounds.” KAVACH performs end‑to‑end speech‑to‑intent: on‑device Whisper‑based ASR (automatic speech recognition) transcribes the voice command; the SLM360 model then parses the text into a structured JSON message (team, grid reference, task, call sign, ammunition type, quantity) which is pushed to the local battle management system (BMS) via a TAK (Team Awareness Kit) plugin. Similarly, queries like “Show me all own‑force logistics vehicles within 5 km of waypoint Bravo” are parsed into a database query and results rendered as map overlays. KAVACH supports English and Hindi voice commands, with code‑switching handled seamlessly.
Security & Zero‑Cloud: All speech‑to‑text and NLP inference runs locally. The SLM360 model file is encrypted at rest and loaded into RAM only after authentication via a hardware security token. No audio or text ever leaves the device, eliminating eavesdropping risk and ensuring functionality in EMCON (emission control) scenarios.
3. SAGAR — Naval Maritime C2 Android App for Indian Navy Operations
Mission: Provide a rugged, offline‑first command and control application for Indian Navy personnel deployed on ships, submarines, and shore stations, enabling tactical coordination, navigational aid, and situational awareness using standard Android devices.
Core Design:
SAGAR is a fully native Android application (Kotlin, Jetpack Compose) hardened for MIL‑STD‑810H environments. It operates entirely without internet or cloud services, relying on onboard ship networks, Wi‑Fi, or mesh radios for inter‑device communication. Key modules include:
-
Tactical Plot (WECDIS‑like): Displays own‑ship position from internal GPS/GNSS, AIS targets (via serial‑to‑Bluetooth adapter), and manually plotted contacts on a high‑performance vector chart (S‑57/S‑63 ENCs loaded from encrypted SD card). Target motion analysis (bearing‑only TMA) calculations run locally.
-
Communication Console: Encrypted text chat and file transfer over LAN with priority‑based message handling. Pre‑formatted naval message templates (e.g., OPGEN, SITREP) are fillable offline.
-
Navigation & Weather: Onboard celestial almanac, tidal prediction engine, and importing of GRIB weather files from ship’s satcom terminal (but processed offline). Route planning with automatic great‑circle/rhumb‑line calculations.
-
Resource Board: Tracks crew watch bills, ammunition states, fuel, and provisions — all synced peer‑to‑peer when devices connect briefly.
Security & Resilience: The app’s APK is signed with a defence‑grade certificate and uses Android’s hardware‑backed Keystore for encryption. Data at rest is AES‑256 encrypted. All features function fully in airplane mode; synchronization between devices occurs via a store‑and‑forward mechanism over ad‑hoc Wi‑Fi or serial link, ensuring no dependency on any external server. SAGAR has been successfully deployed in multiple Indian Navy exercises, providing a low‑cost, readily available C2 terminal that complements existing combat management systems.
4. Data Nexus — Six‑Layer Entity Network Intelligence Platform
Mission: Fuse, visualize, and analyse multi‑domain intelligence on individuals, organisations, locations, events, communications, and documents to reveal hidden networks and support targeting decisions — all on a self‑contained, air‑gapped workstation.
Core Concept:
Data Nexus models the battlespace as an interconnected graph across six distinct yet interlinked layers:
-
Personas — Individuals (known and unknown), with biometrics, aliases, and behavioural signatures.
-
Organisations — Military units, militant groups, political factions, with hierarchical relationships.
-
Locations — Geospatial points of interest (bases, safe houses, meeting points), with temporal activity patterns.
-
Events — Operations, meetings, attacks, convoys, with timestamp and geotag.
-
Communications — Phone calls, messages, radio intercepts, with metadata and transcriptions.
-
Documents — Seized papers, digital files, reports, with text extraction and entity linking.
Platform Functionality:
All data ingestion is manual or via tactical feeds (USB, radio datalink) — no cloud connection. A graph database (SQLite with a custom graph engine or JanusGraph embedded) stores entities and relationships. A local AI module performs entity resolution (deduplicating “Mohammed Khan” across sources using fuzzy matching and context), relationship extraction (from unstructured text using a small on‑device NER model), and link analysis. The operator interacts through a dynamic network visualization interface (2D/3D) that highlights shortest paths, clusters, and betweenness centrality. Temporal and spatial filters allow “travel back” to see network state at a specific date.
Six‑Layer Interrogation:
A user can right‑click a person and instantly see all organisations they belong to, all locations they visited, all events they participated in, all communications they had, and all documents that reference them. The system automatically suggests hidden links — e.g., two seemingly unconnected individuals who frequent the same location at overlapping times or who both called the same intermediary — ranked by probability scores. All analysis runs locally; the platform is packaged as a portable executable with an embedded web server, accessible from a browser on the same machine, requiring no installation or internet. The database itself is encrypted and can be transferred via secure storage.
Suites Integration & Common Zero‑Cloud Architecture
All four tools are designed to be interoperable within a broader tactical ecosystem while remaining independently functional. They share a common secure IPC (Inter‑Process Communication) bus over localhost, allowing cross‑tool data enrichment: an alert from VAJRA (drone detected) can automatically appear as an event in Data Nexus and trigger a voice notification via KAVACH. SAGAR can publish own‑ship position to a Data Nexus layer for multi‑domain common operating picture. The entire suite runs on a hardened Linux or Android baseline, using a lightweight PKI for authentication and AES‑256 for all persistent data. No telemetry, analytics, or diagnostic data is ever transmitted out, ensuring full operational security. This zero‑cloud‑dependency design enables mission success in the most austere and contested environments — from submarine operations in the Indian Ocean to forward posts in high‑altitude warfare.