Track first.Mount anything.

[ THE PROBLEM // COST EXCHANGE ]

Air defence is losing the cost exchange.

• 01 // Cost disparity

  Interceptors costing hundreds of thousands to millions of dollars are fired at drones that cost hundreds to a few thousand.

• 02 // Magazine depth

  Launchers run dry. Swarms don't.

• 03 // Human bottleneck

  One operator per weapon can't keep up with a saturation attack.

• 04 // Monolithic systems

  Each counter-drone system ships as one closed box with one weapon. You can't add a cheaper effector without buying a new system.

[ HOW IT WORKS // THE NETWORK ]

One network, five steps.

One tracking network. Any effector. A human authorises every shot.

1. 01

   Cue

   Radar, RF or acoustic sensors flag a contact. These are existing sensors; we integrate with them.

2. 02

   Track

   The nearest Killswitch nodes slew to the contact. Wide camera, zoom camera, AI detector, predictive tracking. Two nodes on one drone give its 3D position.[Triangulation: target]

3. 03

   Decide

   The network classifies every drone, ranks threats by what is heading for people first, and proposes the cheapest effector for each.

4. 04

   Authorise

   Human in the loop

   An operator approves, edits or rejects the plan. Nothing fires without a human. One operator for about ten nodes.[Target]

5. 05

   Engage and assess

   The node holds aim while the effector works, then confirms the result or re-queues the target.

[ EFFECTORS // BRING YOUR OWN INTERCEPTOR ]

Bring your own interceptor.

The tracker is the product. The weapon is a module.

• Small quadcopters, short range, clear air

  Proposed effector

  Laser module, 5 kW-class[Target], 3–4 km reach[Target]

  Why

  Cost per shot is electricity; no debris from a missile.

• Shahed-class fixed-wing, long range, or rain and fog

  Proposed effector

  Tube-launched interceptor drones, ~US$2,000 class, ~25 km reach[Cited]

  Why

  All-weather; far cheaper than a missile.

• High-tier threats

  Proposed effector

  The defender's existing guns or missiles

  Why

  Saved for what only they can stop.

ReDome: our own concept for a reusable interceptor that returns to base on a miss. Concept stage, not built.[Concept]

Interceptor cost and range refer to Sting-class interceptor drones as reported by Al Jazeera (4 May 2026). Killswitch does not build interceptors or high-power lasers. Effectors are supplied and operated by the defender at licensed sites.

[ SAFETY // HUMAN AUTHORITY ]

Designed so a fault makes it stop, not fire.

1. 01

   The node is the safety authority. Limits, watchdog and e-stop live on the microcontroller. The host can't override them.

   [Built]
2. 02

   Nothing moves until an operator arms it.

   [Built]
3. 03

   Detection is not authority. A single false detection can never move the head; only a confirmed lock can.

   [Built]
4. 04

   Silence means stop. No heartbeat for 500 ms puts the node in fail-safe.

   [Built; verified in simulation]
5. 05

   Every engagement needs operator authorisation, and every proposal, approval and override is logged.

[ PROGRESS // WHAT'S REAL TODAY ]

What we've built, and how we know.

• Detector latency

  41 ms

  [MEASURED]

  Per frame: drone detector on an Apple M1 Pro laptop CPU.

• Tracking error

  0.11°

  [SIM]

  Mean error on a 20°/s crossing target, with 80 ms camera lag and miscalibrated servos. Real firmware logic.

• Automated tests

  235 + 34

  [MEASURED]

  235 automated tests plus 34 firmware test cases. No hardware needed.

• Deadman cut-off

  254 ms

  [MEASURED, SEP 2026]

  Deadman response, with a 2,002 ms burn ceiling, on the hackathon rig.

• Firmware interlocks

  13/13

  [MEASURED, SDTH RIG FIRMWARE V2]

  Interlock checks passed on the SDTH rig, firmware v2. Not re-run on later firmware.

• ATAK-CIV tracks

  Live

  [DEMONSTRATED, SDTH]

  Live tracks sent to ATAK-CIV from a real node plus simulated nodes.

Built vs planned.

• Single-node tracking software, lock gate, safety nodeBuilt, verified in simulation
• Physical velocity tracking rigParts in hand, assembly under way
• Zoom camera, XY head, vehicle nodeNext phase
• Multi-node triangulation and effector allocationPlanned
• Laser and interceptor integrationWith defence partners, at licensed ranges

[ ROADMAP ]

From bench rig to partner pilot.

1. Phase 01 // Current

   Software MVP

   Tracking software and bench rig; tracking a target through 360°.[Target]

2. Phase 02

   Hardware MVP

   Zoom camera, XY head and a vehicle-mounted node; long-range tracking trials.

3. Phase 03

   Pilot

   A second node, networked tracking, and a trial with a defence partner.

[ SIMULATION & EFFECT ]

Imbalance over vaporisation.

A 5 kW-class beam[Target] heats a weak structural point until it warps: 5.2 s to effect on a Shahed-class airframe[SIM, transient thermal model]. Not field-tested.

• The model

  Shahed-class

  Transient thermal simulation of a Shahed-class airframe.

• The beam

  5 kW-class[Target]

  Heat concentrated on a weak structural point, not the whole airframe.

• Time-to-effect

  5.2 s[SIM]

  The composite spar warps in the model. Not field-tested.

[ OUR TEAM ]

Built by engineers from NTU, NUS and industry.

• 

  Rohit Panda

  AI Engineer

  Tracking software

  MSc, NTU

• 

  Aadith Yadav Govindarajan

  Material Science Engineer

  Thermal simulation and rig integration

  I-FIM, SafeRail.AI

• 

  Thejus Aravind

  Mechanical Engineer (aeronautical)

  Gimbal and mounts

  NUS Mars Rover

• 

  Evan Tok

  Electrical Power Engineer

  Power and safety electronics

  Siemens Energy

[ BRIEFING // REQUEST ACCESS ]

Request a briefing.

For defence organisations, primes, investors and grant evaluators.