Vertically integrated physical AI, not a gig app
We design the units, write the dispatcher and intend to operate the fleet ourselves, rather than assemble a marketplace on top of somebody else’s. The same stack carries a second line: industrial robots for plants that cannot hire, cannot hold a tolerance late in a shift, and cannot staff the night.
Rider-only delivery has hit its ceiling.
Cost per parcel has stopped falling, congestion keeps rising, and reliability moves with weather and shift supply. Adding riders adds cost linearly and adds nothing structural.
The defensible layer is the dispatcher.
Cost per parcel is not set by the vehicle. It is set by how much of a trip that parcel shares. Whoever can mix modes and mix urgency — batching a planned parcel into a route already running, and pricing the interruption when it cannot — sets the cost floor.
Handoff infrastructure first, machines second.
Most attempts at this start with the robot. We start with the places a robot can put a parcel down — lockers, ports, swap cabinets — because without them a machine leg cannot exist in an Indian city at all. That layer is also the cheapest and least regulated part to build.
Four curves crossing at once
Quick commerce has trained Indian cities to expect delivery in minutes, in tier-2 cities as well as metros.
Drone rules, certification pathways and green-zone frameworks now exist, so an operator can plan against them.
Motors, packs, compute and sensing have fallen far enough that a locally built rover is a capex line, not a research project.
Made-in-India incentives and a deep local supply chain make in-house fabrication cheaper than importing units.
Cost per parcel falls when a trip is shared, not when a rider is added
Adding riders adds cost in a straight line. Adding notice does not — it adds density to trips that were already being made. This is the whole reason the dispatcher is the asset and the vehicles are not.
Your parcel gets a trip to itself, and you are paying for the interruption as well as the distance. Worth it when a parcel genuinely cannot wait — and wasted when it can.
A rider-only network optimises one variable: which rider. The dispatcher optimises mode and urgency together — bicycle through to truck, rover and drone, against now, batched or planned.
A parcel booked a day out is not a delivery, it is capacity to fill a trip that is already being made. The more of the book that is planned, the higher the drops per trip, and the lower the cost of every parcel on it.
Urgency is charged for rather than averaged across everyone. Customers who can wait fund the density; customers who cannot pay for the interruption they cause. Most networks quietly do the opposite.
Batching does not stop working when the crew member is a rover or a drone. A machine on a planned route is the cheapest leg in the network, which is why the handoff layer is worth building before the machines are.
Industrial and construction robots, on the same stack
Chassis fabrication, perception on unstructured ground, remote assist and running machines without a person on each one — a delivery network makes us solve all of it against a paying load. The same stack sells into a demand problem that already exists: plants cannot hire the people they need, cannot hold a tolerance late in a shift, and cannot staff the third shift at all. A humanoid and an autonomous carrier answer those. Two product lines, one engineering effort.
The work a person would do, on the shift nobody can staff
Material across the floor, driverless, to the millimetre
- 2026 Q4 · TARGETRider network opens on the first routes, with the brands who sign up to them.
- 2026 Q4 – 2027 Q1First drop boxes installed; brand API and dashboard hardened.
- 2027 H2First rovers on our own estate, remote-assisted, geofenced.
- 2028 ONWARDSAir legs after type certification; second city once Vadodara holds SLA.
Numbers, unit economics and the build plan
Sent to institutional and angel investors on request. We do not publish revenue or funding figures on the site.
INVESTOR@THECREW.ONE