Sovereign counter-drone defense · India

When the jammer fails, something has to hit it.

tratr is an AI-directed, close-in point-defense system that physically defeats the drones nobody else can stop — RF-silent FPV and fiber-optic attackers (radio-silent; no signal to detect or jam) that fly straight through jamming.

See it, classify it, hit it — while friendly aircraft are recognised and left alone.

Book an intro call → ▶Watch it track a live droneupdate
Air-gapped edge AI LLM out of the kill loop Threat classification — drone vs bird / aircraft Standalone-first · grid-ready
LIVE: AUTONOMOUS DETECT → CLASSIFY → ENGAGE
01 — The threat

The drone that ignores your defenses.

India's counter-drone stack is overwhelmingly RF detection and jamming. That entire layer goes blind against fiber-optic-tethered and fully autonomous FPV drones — no radio to detect, nothing to jam. These are the weapons redefining the modern frontline, and they arrive RF-silent.

tratr attacks the white space: passive detection — acoustic plus electro-optical vision — paired with cheap kinetic defeat. You don't need to hear a radio if you can see it, hear it, and hit it.

RF-silentFiber-optic & autonomous FPV
PassiveNo emissions to detect us
KineticPhysical defeat, low cost/shot
Legacy RF jamming2 / 3 missed
RF drone
Fiber-optic
Autonomous
tratr · passive + kinetic3 / 3 · design target
RF drone
Fiber-optic
Autonomous
Passive sensing sees the drone regardless of its radio — every contact is detected, classified as friend or threat, and handed to fire-control for kinetic defeat.
The benchmark
The model's proven.

The clearest proof this category is real is 9 Mothers (Austin, YC 2026) — AI shotgun turrets for close-in drone defense, already delivering to U.S. forces. But as a U.S., ITAR-controlled system, any India sale needs case-by-case U.S. export approval — and can never be sovereign or free of a foreign kill-switch, colliding with India's indigenous-procurement mandate. The playbook is proven; the Indian market is effectively open.

The opening
A sovereign answer, built in India.

India needs this capability indigenous, air-gapped (fully offline), and free of foreign kill-switches. tratr builds the Indian equivalent from open-weight AI on-site — no cloud, no external calls, provenance-controlled. The beachhead: BSF → Navy → Army Air Defence.

02 — The system

Detect → Track → Aim → Engage.

A deterministic sensor-to-effector (the weapon) pipeline — from detection to the trigger — running entirely on a hardened edge node (an on-site computer). No generative AI anywhere in the firing chain.

→
STAGE 01

Sensors

Passive acoustic direction-finding fused with EO/camera vision — catches the fiber-optic and autonomous drones RF systems can't, and emits nothing itself.

  • Mic-array direction-of-arrival (DoA)
  • Open-weight EO detector (RF-DETR / D-FINE)
  • Threat classification: drone vs bird / aircraft
→
STAGE 02

Fire-Control

Detections become tracks; tracks become an aim solution; the solution becomes pan-tilt commands. Classical, low-latency, testable.

  • Multi-object tracking
  • Aim-solution solver
  • Deterministic — no LLM in the loop
→
STAGE 03

C2 & Edge

Command & control runs on an on-site edge GPU node — offline, no external calls. Operator sees live tracks; an air-gapped LLM writes after-action reports only, never firing.

  • Live track / alert / log dashboard
  • Secure boot + disk encryption
  • Signed offline update path
  • Documented track-export + cue-in interface (ICD — our published integration spec)
STAGE 04

Effector

A safe stand-in (laser / airsoft) proves the full loop today; a 12-gauge shotgun effector follows post-licence. Same interface — only the muzzle changes.

  • Safe stand-in for bench + demo
  • 12-ga shotgun (post-licence)
  • Faster slew + belt-fed later
LLM out of the kill loop Detect → track → aim → engage is fully deterministic CV/DSP (computer-vision & signal-processing — rule-based, not AI guesswork). The language model lives only in the human / analysis layer — a safety property, engineered in, not an afterthought.
03 — Where we are

The loop is closed — the turret follows a live drone.

The two tracks are now wired together: the targeting brain (vision + fire-control) drives the motion platform (the test turret) automatically. On the workbench it detects, tracks and follows a hovering drone end to end — camera in, servo commands out, no human in between.

TRL 4 · integrated system validated in the lab — detector, tracker, fire-control and turret running as one loop TRL 5 next · relevant environment — outdoors, drone in flight at range
Track A

Targeting brain

Driving the turret

Open-weight detector + multi-object tracker running live on the bench camera and issuing aim commands to the turret — closed-loop on a screen target, then on a real hovering drone. Every run is logged and replayed against held-out clips before any change ships. Now hardening the detector against harder domains — small/distant targets, clutter, false-alarm sources.

Detection pipelineOperational
Tracking & aim engineClosed-loop on the bench
Outdoor tracking at range — drone in flightStarting
◆ Open-weight, Apache-2.0 (open licence)◆ Live bench inference◆ Live-drone tracking (bench)◆ Own dataset growing
NowHardening the detector: small/distant targets, clutter, and the false-alarm sources the bench has surfaced — with held-out replays as the gate for every change.
Track B

Motion platform

Closed loop · bench

Test turret on the workbench — COTS (commercial off-the-shelf) pan-tilt with smart servos, a safe stand-in effector (laser), and a hardware e-stop. The fire-control loop takes aim commands straight from Track A and the turret follows the target; servo-side travel limits and an arm-time self-test bound every run. Servo latency, pointing repeatability and loop rate are measured, not assumed.

Pan-tilt (aiming) control loopClosed
Closed-loop aim (A→B)Bench-proven
Fire channel — arm · interlocks · engageIn build
◆ COTS gimbal + smart servos◆ Safe stand-in effector◆ E-stop + servo-side limits
NowMeasuring the latency budget end to end (camera → detector → servo) and tightening the loop rate; safety fences for laser-on runs with people in the room.
  tratr-edge · closed-loop bench · sensor→effector
First live-drone run · workbench · 16 Sep 2026

Watch the turret follow a hovering drone.

Camera in, servo commands out. Detector → tracker → aim → pan-tilt, running on the bench with no one steering. Short clip, unedited apart from the cut.

Closed loop · TRL 4Safe stand-in effectorIndoor · short range
tratr test turret following a hovering drone — bench run

Bench prototype, indoors, at short range, with the safe stand-in effector. This shows the plumbing closes end to end. No performance figures are claimed from it — field-validated metrics are gated on held-out, real-world testing and reported at pilot.

Test-bench footage shown for capability illustration. Field-validated performance metrics are gated on held-out (unseen), real-world testing — reported honestly at pilot.

04 — Operational roadmap

Software on-ramp → licence → kinetic.

Each phase de-risks the next. Traction and end-user demand are built before the hard walls — the arms licence and the kinetic build — are approached.

P0 · FoundationJul – Sep 2026Current
Software targeting brain wired to a safe test turret — closed loop demonstrated on the workbench, following a live drone; BSF discovery (underway).
Ships
Detect→track→aim on the workbench ✓ · live-drone bench run ✓ · TRL 4 ✓ · BSF talks
Gate to next
Closed-loop demo ✓ · ₹50L secured · BSF interest
Funding
Bootstrapped
P1 · Traction & LicenceOct 2026 – Mar 2027Next
Deployable software product, an iDEX (govt defence-innovation programme) / BSF pilot, the end-user requirement letter, and the arms licence filed.
Ships
Hardened air-gapped build · operator C2 · pilot deployment · TRL 5–6
Gate to next
Live pilot + end-user letter + licence filed
Funding
Non-dilutive govt. grants (iDEX/ADITI) or pre-seed ₹1–3 Cr
P2 · Kinetic PoCMar – Dec 2027Planned
Kinetic proof-of-concept (live-fire) on a range, a fieldable turret, and the first order or LOI.
Ships
12-ga effector integration · range live-fire PoC · fieldable unit
Gate to next
PoC defeats a drone + licence + order/LOI
Funding
Seed ₹15–40 Cr (estimated)
P3 · Product line2028 +Planned
Faster slew platform, belt-fed variant, counter-Shahed interceptor (against large "kamikaze" attack drones), ammunition via partner.
Ships
High-speed platform · belt-fed · counter-Shahed line
Gate to next
Scaled deployments & repeat orders
Funding
Series A
05 — The vision

India's sovereign close-in air defense.

Software is the on-ramp, not the destination. It makes tratr a legitimate defense entity, wins end-user backing, and turns the arms licence from a cold application into a demand-backed one — funding the kinetic build with proof, not promises.

01

Software brain

AI detection + fire-control + C2 (command & control) establish tratr as a real defense vendor.

02

End-user backing

BSF / Navy demand + the end-user letter unlock the licence path.

03

Proven capability

Mechatronics + targeting brain prove we can build the effector (the weapon).

04

Arms licence

Approached from traction & demand — not from zero.

05

Kinetic product

The turret, then a full point-defense product line.

01

Proven, not research

Integrate proven parts — open CV (computer vision), COTS (off-the-shelf) gimbals, standard effectors. No science-project gates on the critical path.

02

Sovereign air-gapped edge

On-site compute, open-weight models, no cloud, no external calls. Air-gapped (fully offline); defense-in-depth by design.

03

LLM out of the kill loop

Detect→track→aim→engage is deterministic CV/DSP. The LLM (AI language model) lives only in the human / analysis layer.

04

Capital-light until traction

Raise on proof. Non-dilutive grants (iDEX / ADITI); equity only where it accelerates.

06 — Why this is winnable
A capability that doesn't exist here

tratr defeats the RF-silent drones today's jam-based systems miss — passive detection paired with cheap kinetic kill. No fielded Indian system does this. And it's open where every incumbent is closed: a documented track interface, designed to plug into Akashteer/SAKSHAM-class C2 grids (India's air-defense command networks) — standalone-first where no grid exists.

The scarce asset

Warm BSF relationships and prospective Navy access — the procurement door and end-user letter most defense startups never reach.

Timing

The threat is here now, the U.S. benchmark is structurally locked out of India's sovereign-procurement lane, and non-dilutive defense capital is actively deploying.

Raising pre-seed

Back the sovereign answer to the drone threat.

We're applying for non-dilutive government grants (iDEX/ADITI) and raising a pre-seed round (₹1–3 Cr) — to take the closed-loop system from bench to a BSF pilot and file the licence. Warm intros welcome.

Grants + pre-seed ₹1–3 Cr · defense-tech · India
Book an intro call →