LiftSense fuses load, vibration, wire-rope and thermal sensors on your lifting machinery with computer vision and long-context AI reasoning, so a registered Lifting Machinery Inspector arrives at every DMR 18 thorough examination already knowing what to check first, and leaves with a defensible, standards-aligned report in minutes, not hours.
Live today across pilot sites in Gauteng and the Western Cape; the roadmap extends coverage nationally as more LMEs onboard their fleets.
Registered Lifting Machinery Inspectors are personally accountable for every thorough examination they sign, yet they're still working from clipboards, torque wrenches, and the naked eye against thousands of load cycles they never see.
A single component (e.g. one sling, one bearing) is degraded. AI isolates it; the rest of the machine stays in service pending repair.
Correlated readings across multiple machines of the same model/batch suggest a design or maintenance-programme problem worth escalating fleet-wide.
Sensor data is ambiguous or conflicting. The system explicitly withholds a finding and flags the item for manual, in-person inspection.
In every case, LiftSense produces a lead for a registered LMI to investigate — never a certified determination. Only a human inspector signs a thorough-examination report.
Every stage produces an artifact a human can audit — a raw reading, a confidence score, a draft recommendation, a signature. There is no step where a machine acts on its own.
Turns time-series load, vibration and thermal data into a remaining-safe-life estimate per component. Runs on AWS SageMaker.
Converts a prediction or defect finding into a short, plain-language recommendation with its reasoning shown. Runs on a current GPT model.
Checks a new finding against years of that machine's prior thorough-examination reports for precedent. Runs on a current Claude model, chosen for its long context window.
Scans LMI-captured photos of hooks, welds and wire rope for cracks, corrosion and strand damage. Runs on Google Vertex AI custom vision.
Produces a simple annotated diagram attached to the report, explicitly labelled AI-generated, showing where a defect was found.
The mandatory final decision-maker, registered with ECSA. No finding becomes a report without their signature. Always present, never optional.
Model identifiers above should be verified against each provider's current documentation rather than assumed from any static spec — providers update and rename models frequently. This build's AI outputs are simulated to demonstrate the workflow shape rather than live model calls, and running the full stack at fleet scale is estimated in the low tens of US cents per inspection for LLM/vision calls, excluding IoT connectivity and storage — a rough, labelled range, not a quote.
This model is simplified for clarity. The real system follows the same shape — sensor → AI pipeline → human decision — using real load cells, rope sensors and camera feeds on actual machinery.
Cranes and hoists move tonnes of load over people's heads every day. When something is missed, the cost is measured in lives, not just downtime — which is exactly why South African law makes a registered human inspector, not a machine, the final authority.
Reports that U.S. crane-related fatalities averaged around 42 per year between 2011–2017, down from about 78/year in the 1992–2010 period, based on BLS Census of Fatal Occupational Injuries data.
Notes that dropped loads from overloading or poor rigging, and being struck by an object, are among the most common causes of fatal crane incidents.
Defines a Lifting Machinery Inspector as a person registered with the Engineering Council of South Africa, and lists the SANS standards (e.g. SANS 4309 for wire rope) that govern thorough examinations.
Requires that steel-wire ropes maintain a safety factor of at least five relative to the machine's safe working load.
Explains that a lifting machine must undergo a thorough examination at intervals not exceeding 12 months, and lifting tackle at intervals not exceeding three months.
Describes how the 2005 gazette required Lifting Machinery Entities to register their load-test personnel as Lifting Machinery Inspectors through ECSA.
LIVE — this panel fetches real current wind and temperature data from Open-Meteo every 5 minutes for each pilot site. The "stage" label under each site name is our own internal rollout designation, not derived from this feed.
"I've spent enough time on and around lifting equipment to know that the gap between a routine lift and a serious incident is often just a detail nobody was watching closely enough. Karabokhotso Projects exists to close that gap — to give the people who inspect and operate this machinery better information, sooner, so their judgement is backed by real data instead of guesswork. What we're building isn't meant to replace the inspector's sign-off; it's meant to make sure that when they sign, they've seen everything there was to see. That's the standard I want this company held to."
Demo access only. No account is created or verified. All data shown after entry is simulated for demonstration purposes.
Fires the full pipeline for RD-014 end-to-end, and pauses for a human decision before anything is issued.