The inspection report writes itself. The expert still signs it.
DRISHTI-AR puts computer vision and augmented reality in the hands of maintenance crews working on transport-aircraft fleets. AI finds candidate defects in the camera frame; a human expert adjudicates every one; and the sign-off lands in an append-only audit trail. Speed from the machine, accountability from the person.
- [ YOLOV8 + CLASSICAL CV ]
- [ MACRO F1 ~77% ON REAL DATA ]
- [ APPEND-ONLY AUDIT TRAIL ]
DRISHTI-AR AR + AI AIRCRAFT MAINTENANCE PLATFORM
Corrosion does not wait for the next scheduled deep inspection. DRISHTI-AR turns every walk-around into a data point: live camera capture, instant analysis, expert review, and a fleet-level fitness picture that improves with every frame.
Dual-path defect detection
YOLOv8 deep learning fine-tuned on real corrosion imagery detects corrosion, cracks, and dents — macro F1 of roughly 77% on held-out real test data, with 100% F1 on the dent class. A classical CV path independently checks fasteners and coatings.
Predictive maintenance
Weibull hazard modelling per airframe zone converts inspection history into a fitness score and an estimated remaining useful life in days — a number a fleet manager can plan against.
Roles, auth, audit
Role-based access for technician, expert, engineer, and admin; JWT authentication; and an append-only audit trail behind every decision the platform records.
Sign-off you can file
Branded PDF inspection sign-off reports and fleet fitness reports, generated from adjudicated findings — documentation as a by-product of the work, not a second job.
AR-guided procedure
Augmented overlays walk technicians through inspection and maintenance steps in place, on the airframe, with the analysis happening live as the camera moves.
Beyond the hangar
The same platform extends to immersive training, digital-twin visualisation, and XR for engineering education — one pipeline from capture to classroom.
Immersive systems as engineering tools
We treat AR and VR the way we treat cryptography: as instruments, not effects. An overlay earns its place by reducing error rates; a digital twin earns its place by answering questions the physical asset cannot.
Immersive training lets crews rehearse procedures on equipment that is too expensive, too remote, or too critical to practise on. Digital-twin visualisation puts inspection history, fitness scores, and predicted degradation onto the geometry of the asset itself. XR for engineering education takes the same rigour to the lab bench and the classroom.
Rehearse before you touch
Procedure walk-throughs in headset, scored and repeatable, with the same adjudication discipline as live inspection.
The asset, annotated
Fleet fitness and zone-level hazard data rendered on the airframe geometry — degradation you can walk around.
XR for engineers in training
Interactive systems teaching inspection technique and failure recognition long before the first real airframe.
Questions we actually get asked
Does the AI replace the inspector?
No, and it is not designed to. The AI proposes candidate defects; a qualified expert adjudicates every finding before it enters the record. What changes is coverage and speed — the expert reviews everything the camera saw, not just what a tired eye happened to catch.
How good is the detection, honestly?
Macro F1 of approximately 77% across corrosion, crack, and dent classes on held-out real test data — not synthetic imagery — and 100% F1 on the dent class. We publish the measured numbers because an adjudication workflow only works when you know exactly how much to trust the machine.
What makes a report "audit-grade"?
Every capture, AI finding, adjudication decision, and sign-off is written to an append-only audit trail under role-based access and JWT authentication. The PDF report is a rendering of that trail — nothing in it exists without a recorded decision behind it.
Where does the remaining-useful-life number come from?
From Weibull hazard modelling fitted per airframe zone using the platform's inspection history. The output is a fitness score and an estimated remaining useful life in days — with the model and its inputs inspectable, because a prediction you cannot interrogate is just a guess with confidence.
Bring a photograph. Leave with a finding.
A DRISHTI-AR evaluation starts with your own imagery run through the live pipeline — detection, adjudication, and a signed report, end to end.