The Hardware Trap & The AI Abstraction.
Incumbents sell better antennas. Kinetic sells chainage-locked fouling intelligence and CapEx targeting — on whatever hardware you already run.
Jesse James
An iPurpose Company
gprkinetic.pro
The Incumbents: 2026-2030
The global Ground Penetrating Radar market is experiencing robust expansion, but the innovation curve is entirely hardware-focused. The top 9 market leaders are locked in a race to improve antenna frequencies and signal-to-noise ratios. They provide excellent raw data (B-Scans), but push the most expensive bottleneck—data interpretation—onto the end user.
GSSI
Pioneering hardware, but software remains tied to manual interpretation workflows.
Leica Geosystems
Premium precision measurement, focused on traditional GIS rather than predictive AI.
Radiodetection
Strong in utility location, limited in automated multi-layer rail ballast analysis.
IDS GeoRadar
Excellent multi-frequency arrays; their hardware acts as our ideal data acquisition layer.
Sensors & Software
Reliable equipment, but visualizations still require expert geophysicist review.
MALA Geoscience
Flexible systems, yet lacking edge-compute capabilities for real-time anomaly detection.
Applus+
Massive service footprint, heavily reliant on expensive human-capital for inspections.
Geoscanners
Specialized high-frequency tools, but disconnected from cloud-native fleet management.
Utsi Electronics
Precision instrumentation for academics, hard to scale for enterprise rail networks.
Disruption Phase I: Deep Learning Inversion
Legacy software requires a human geophysicist to manually trace hyperbolas on a black-and-white radargram. Kinetic bypasses this entirely using advanced deep learning architectures for real-time permittivity inversion.
Inspired by recent advances in subsurface root detection, Kinetic utilizes pyramidal convolutions coupled with Vision Transformers. This allows the model to simultaneously extract macro-features (large voids) and micro-details (fine ballast fouling and moisture gradients) across heterogeneous soil layers.
Instead of relying on computationally expensive iterative FWI algorithms that fall into local minima, our neural networks directly output the spatial distribution of permittivity from the B-scan data in milliseconds, mapping complex underground structures accurately without human intervention.
Research track: multi-sensor fusion
Pilot/research — not live in product today. Fusing subsurface GPR with surface context (vision, mmWave) for chainage-locked anomaly correlation on hi-rail collection.
RESEARCH · illustrative roadmap
Dynamic Vision Sensors (DVS)
Unlike frame-based cameras that suffer from motion blur at 100 km/h, event-based neuromorphic cameras capture pixel-level brightness changes at microsecond resolution. When mounted on hi-rail vehicles, they provide perfectly crisp surface-defect mapping.
Recurrent Vision Transformers
Fusing 77GHz mmWave surface radar with DVS data and feeding it through Recurrent Vision Transformers (RVT) allows our platform to correlate surface anomalies (cracked ties, loose fasteners) with subsurface GPR anomalies (ballast fouling) automatically.
Hardware-agnostic intelligence layer
Ingest IDS, ENSCO, or GSSI outputs — add Res-SAM, ontological verification, and Economic Oracle framing without swapping your collection fleet.
Partner path
See the platform, then scope a corridor
Diagnostic twin → real GPR example → pilot briefing. Open demos — no login wall.