13 CLASSES
13 Fine-Grained Defect Classes
Hollowing, cracks, spalling, seepage, aging and more — read with a recall-first strategy: better to over-flag for professional review than to miss real risk.
— MOAT · Diagnostic Capabilities
We fly against Taiwan's glass curtain walls every day.
Diagnostic Capabilities
So we see deterioration others can't see, and hit problems others never hit. The former is the business; the latter is the R&D — they are the same thing.
#inspek AI#GDS-Pilot#Data platform
13
Fine-grained defect classes
68,000+
Taiwan facade training images
±2°C
Thermal accuracy
cm-grade
RTK defect positioning
— 001 · Image Interpretation
inspek — In-House Defect-Reading AI
Not a generic model. inspek is trained on more than 68,000 images of real Taiwanese building facades and reads 13 fine-grained defect classes — because Taiwan's tiles, pebble-wash, and curtain walls don't look like the walls in academic datasets.
DUAL-AI
Dual-Model Cross-Reading
A segmentation model and a multimodal large model cross-vote to reduce single-model misreads; every defect carries a confidence grade for professional sign-off.
THERMAL-FUSE
Thermal-Visible Geometric Fusion
Infrared thermal imagery is geometrically aligned with visible light, so ΔT anomalies map directly onto facade coordinates — hollowing isn't 'somewhere over there', it has coordinates.
MATERIAL
Material-Aware Interpretation
Eight facade materials modeled separately — tile, stone, pebble-wash, curtain wall and more. The same shadow means hollowing on tile and may mean nothing but grime on stone.
— 002 · In-House Fleet
GDS-Pilot — In-House Flight Control & Wall-Proximity Aircraft
Off-the-shelf drones were never designed to fly against a glass curtain wall while fighting nozzle recoil. So we forked PX4 to build GDS-Pilot and construct our own wall-proximity aircraft.
PX4 FORK
GDS-Pilot Flight Control
A PX4 fork tuned for wall-proximity work: wind shear, multipath interference, and nozzle-recoil compensation — problems that don't appear in open-source default parameters.
RTK
Centimeter-Grade Positioning
RTK centimeter-grade positioning plus laser ranging pin every image and every defect onto the facade coordinate system, enabling precise same-path re-flights year over year.
SENSORS
Multi-Sensor Payload
48MP triple-lens visible camera + VOx 640×512 thermal imager (±2°C) — one flight captures both the visible and thermal data layers.
FIELD
In the Field Every Day
Our in-house pilot fleet works facades, solar farms, and industrial sites daily. The failure modes we design around were encountered in the field, not imagined.
— 003 · Data Platform
Your Building's Medical Record
Fifty thousand square meters of accumulated operational data — every flight feeds the model. Earned with time, not bought with money.
SCORE
Health Score
Every building gets a 0–100 composite score across five dimensions — hollowing, cracks, spalling, seepage, aging — tracked year over year.
GEO-TAGGED
Coordinate-Tagged Defect Library
Every defect carries facade coordinates, a severity grade, and visible-plus-thermal evidence — auditable, reviewable, and directly actionable by contractors.
TIMELINE
Year-over-Year Timeline
The same wall compared across years: deterioration rate, spread direction, risk curve. Problems aren't discovered; they're predicted.
FORECAST
Budget Forecasting
Trend data projects five years of maintenance needs and budgets — turning facade upkeep from a surprise expense into a plannable item.