Dhakma Edge On-device AI analysis at the point of action. · Real-time. Zero cloud. 91.4% accuracy. Need on-premise data AI? → Core
Dhakma Edge · On-Device AI

See everything.
Upload nothing.

Real-time AI analysis. On-device. No cloud required.

PLATFORM: RPi 5 + HAILO-8L
ANALYSIS: ON-DEVICE · REAL-TIME
CLOUD: NONE
ACCURACY: 91.4%
TRACKER: ByteTrack
TILING: SAHI
COMMS: MQTT
CORE SYNC: ACTIVE
Get in Touch Add Dhakma Core →
26TOPS · Hailo-8L
30fpsOn-Device
91.4%Detection Accuracy
Real-time on-device AI · zero cloud Site intelligence · construction · mining · agriculture ByteTrack persistent object IDs SAHI tiling · long-range small object detection Site Intelligence · Drone Detection · Autonomous Navigation Real-time on-device AI · zero cloud Site intelligence · construction · mining · agriculture ByteTrack persistent object IDs SAHI tiling · long-range small object detection Site Intelligence · Drone Detection · Autonomous Navigation
The Land-and-Expand Dynamic

Edge devices enter at low procurement friction. Once deployed, they generate operational data, detections, and edge cases at scale. AI models degrade as the physical world changes. The answer is Dhakma Core — private retraining, local device management, closed-loop intelligence. Edge creates the requirement. Core solves it.

Platform Specifications

26TOPS · Hailo-8L NPU
30fpsReal-time AI analysis
91.4%Detection accuracy
$0Cloud dependency
HardwareRaspberry Pi 5 + Hailo-8L
Object TrackingByteTrack · Persistent IDs
Small Object DetectionSAHI Tiling · Long-range
Comms ProtocolMQTT pub/sub · Core sync
Degraded CommsGraceful fallback · Full autonomy
Hardware AbstractionHailo · Jetson · NPU-agnostic
Core IntegrationDevice mgmt · AI updates · Performance data
DeploymentRemote standard · On-site available
Edge → Core Expansion

Edge devices generate operational data that AI models cannot process indefinitely. When your environment evolves, your AI must adapt. Dhakma Core closes the loop — private retraining on your hardware. No cloud required.

Learn about Core →

Three Deployable Verticals

Install only
what you need.

Each vertical is independently deployable on the same Raspberry Pi 5 + Hailo-8L hardware. Add verticals as your operational requirements evolve — or activate all three from day one.

01
Site Intelligence
Automated Monitoring & Analysis

Instance segmentation on drone and fixed-camera imagery with per-object mask overlays, volumetric calculations from masked regions, and automated site analysis — all processed on-device. Survey data stays on your hardware. Progress tracking, stockpile measurement, and change detection without uploading a single frame to the cloud.

InstanceSegmentation model
MaskOverlay mapping
VolumetricCalculations
On-deviceProcessing
Construction Mining & Extraction Agriculture Insurance Environmental Survey Property Management
02
Drone Detection
Automated Airspace Monitoring

4-class detection across fixed-wing, multicopter, VTOL, and bird. Classification with confidence scoring, 360-degree sensor coverage with overlapping fields of view, track fusion across multiple devices, and real-time alerting. Proven at 91.4% accuracy on live datasets with advanced tiling for long-range small-object detection.

4Detection classes
360°Sensor coverage
91.4%Accuracy
Real-timeAlert pipeline
Airports & Airspace Event Venues Corporate Campuses Critical Infrastructure Private Estates Defense & Military
03
Autonomous Navigation
Visual-First Drone Navigation

Vision-based approach vector computation with GPS fallback and seamless switching. Full autonomy state machine with comprehensive failsafe — mode verification, heartbeat monitoring, and graceful degradation. Operates independently in GPS-denied or connectivity-limited environments.

VisualPrimary navigation
GPSAutomatic fallback
FullFailsafe coverage
ZeroUplink dependency
Inspection & Survey Autonomous Delivery Agriculture Logistics Drone OEMs Defense Integrators

Edge + Core · The Complete Pipeline

Edge acts.
Core learns.

Dhakma Edge is the field half of a closed-loop pipeline. Dhakma Core is the command layer — training new AI models on your operational data, managing your device fleet, and pushing updates without a cloud dependency at any stage.

Most Edge deployments become Core deployments. Here is why.

Explore Dhakma Core Get in Touch
01
Edge Deployment

Edge devices enter at low procurement friction. $500/device hardware, slides under procurement thresholds. Immediate operational value from day one.

02
Operational Data Accumulates

Edge devices generate detections, false positives, and edge cases at scale. New patterns emerge. Site conditions change. The operational world evolves.

03
Models Degrade

AI models trained on yesterday's data become less accurate over time. The environment your Edge devices operate in is no longer the environment the model was built for.

04
Core Closes the Loop

Dhakma Core retrains on your operational data, manages your devices, and pushes updated AI models back to the field. Locally. Continuously. Privately.

Get in Touch

Edge deployments start
with a conversation.

Dhakma Edge is scoped to your site, platform, and operational environment. No standard pricing, no self-serve configuration — every deployment is different.

Tell us what you need. A Dhakma engineer will respond within one business day.

Premium onsite support — a Dhakma engineer deploys to your location

Premium Onsite Support

A Dhakma engineer comes to you for installation, integration, and training — across the contiguous United States. No offshore handoffs. No remote-only setup.

Also need on-premise data AI?

If your organization also handles sensitive documents, financial records, or classified data — Dhakma Core handles that. One vendor. One pipeline. Complete privacy.

Explore Dhakma Core →
Get in Touch

Handled directly by a Dhakma engineer · Not a sales team

Responses within 1 business day · Direct engineer contact · All inquiries strictly confidential
dhakma.ai · Tharindu 'Mackie' Mathew, PhD · Ex-Microsoft Azure