• TIRESIUS The acoustic digital twin.
  • ARGUS Massive Aperture Distributed Mic Array (MADMA).
  • ORPHEUS The library of acoustic AI models.

Every sound on the battlefield is a signature. Ziz deconstructs it — whether to precisely target an incoming FPV at 150 km/hr, track a warfighter peeing in the bushes, or perhaps in the future, diagnose a loose third piston in an Abrams Tank.

§01 · OUR ACOUSTIC DIGITAL TWIN

TIRESIUS.

PROBLEM · THE SIM-TO-REAL GAP + SITE-SPECIFIC MULTIPATH ANSWER · A PHYSICS-EXACT DIGITAL TWIN TRAINED ON YOUR FOB

Sim-to-real-to-sim.
The loop is the moat.

The single greatest operational hazard for AI systems deployed in combat is the sim-to-real gap — networks trained on sterilized datasets that fail on contact with real-world chaos. Ziz neutralizes this hazard through TIRESIUS, our proprietary j-Wave acoustic propagation engine. It is the ultimate acoustic digital twin.

When Ziz executes a contract to protect a specific forward operating base, we don't start by fielding microphones. We load the exact CAD topography — concrete barriers, radar dishes, localized terrain — into TIRESIUS. The engine synthesizes millions of hours of adversary drone signatures cascading through that exact environment. Our denoisers train on the precise multi-path echoes bouncing off the primary hangar. Before a Ziz node is screwed onto a pole in the real world, its neural networks have already lived at that operating base for a million hours.

The nodes then field. They stream ambient acoustics, multipath, wind profiles, and real drone engagements back to TIRESIUS. The twin re-calibrates. The next generation of detectors trains inside the updated sim. They ship to the field. The loop compounds.

Hardware can be cloned. Algorithms can be copied. A competitor cannot replicate twenty deployments of calibration data without deploying twenty arrays themselves. The sim is the moat. The sensor network feeding it is the only way in.

01 · FIELD 02 · INGEST 03 · TIRESIUS 04 · RETRAIN SIM–REAL –SIM LOOP
SIM → REAL → SIM · COMPOUNDING
j-Wave
Physics-exact acoustic twin.
GPU wave propagation across atmospheric gradients, terrain, and adversary hardware. 10,000 simulated hours per day per DGX. Sim-to-real F1 delta: 2.1%.
EnKF
Continuous data assimilation.
The atmosphere is a chaotic fluid. Our Ensemble Kalman Filter fuses live SIRT inversions against prior numerical forecast states — a massive temporal anchor. Graceful degradation up to 40% node loss. Recovery after blackout: < 2 s.
Neural
Tracker
GRU / Transformer, site-calibrated.
10M synthetic drone flights through a digital twin calibrated to the exact FOB. Learns site-specific multipath, wind biases, and occlusion handling before deployment.
Before a Ziz node is screwed onto a pole in the real world, its neural networks have already lived at that operating base for a million hours.
— PROSPECTUS § 3.4 · NEURAL TRACKER
§02 · OUR DISTRIBUTED SENSOR MESH

ARGUS.

Imagine a microphone array that's 500 meters across.
This is MADMA.

Ziz operates as a passive acoustic network — not a single sensor. Drop as many nodes into the terrain as the fight requires. Each node sharpens the fix. Each new node extends the footprint. Every moving sound source in the battlespace — ground or air — converges into a single track.

THE COST STRUCTURE BEHIND IT
INCUMBENT CLASS
One array at $70,000.
FIXED ARRAY · SQUAREHEAD-CLASS
e.g. Squarehead, Boomerang, fixed directional mics
Localizes from a single high-cost array. Accuracy and range are bounded by one piece of hardware at one location. Coverage scales only by buying another $70K unit and erecting another pole.
ZIZ CLASS
Or $300 per node — as many as the fight demands.
DISTRIBUTED ARRAY · PASSIVE MESH
Ziz · passive acoustic mesh
Low-cost passive nodes cooperate via GPS-PPS-synchronized TDoA. Add a node, add accuracy. Add nodes, add coverage. Precision and footprint scale linearly with the number fielded — not with a single hardware budget line.
§03 · OUR LIBRARY OF ACOUSTIC AI MODELS

ORPHEUS.

The physics of outdoor acoustic
localization is hostile.

Three laws of physics stand between a microphone and a grid coordinate: clock drift, refraction, and multipath. A 5 m/s crosswind combined with a 10°C ground temperature differential induces continuous, severe refraction. Sound bends toward the earth downwind, refracts upward upwind. Legacy TDoA systems feed distorted arrival times directly into their neural nets and Kalman filters. The result is mathematically catastrophic.

CLOCK DRIFT
∆t → 34 cm positional error per 1 ms of drift.
REFRACTION
15 m displacement @ 1 km.
MULTIPATH
4–9 m of scrambled arrivals in urban terrain.
Clock drift · NTP
34 cm
1 ms of ad-hoc mesh jitter translates to 34 cm of positional drift — before the network sees a sample.
Refraction @ 1 km
15 m
A simple crosswind induces a 15-meter positional displacement at 1 km. In C-UAS, that is functionally identical to total sensor failure.
Urban multipath
4–9 m
Echoes off concrete barriers, hangars, and terrain scramble legacy trackers that weight every node identically.
Combined @ 1 km
~19 m
The legacy stack concedes 19 meters of error before the neural network sees a single sample. No model survives that.
Competitors treat the atmosphere as an error to absorb. We treat it as a lens to invert.
— PROSPECTUS § 1.3 · CORE PHILOSOPHY
§04 — THE ADVERSARY

The modern UAS defeats every sensor.
Except the one that hears.

Small UAS platforms are engineered to be invisible to the sensors fielded against them. Radar, optics, and SIGINT have each been systematically defeated. Acoustic emission is the one signature the threat cannot engineer away — physics won't let it. Propellers move air. Air carries sound.

RADAR
Low-RCS composite airframes. Sub-bird radar cross-section. A sensor decades of counter-air doctrine was built around — and that the modern UAS was engineered to walk past.
THERMAL · EO
Electric motors run cold. Matte carbon skin. No exhaust plume. Night optics and long-wave IR return nothing the operator can commit on.
RF · SIGINT
A fiber-optic tether carries command and video. Zero radio emissions, zero spectrum footprint. Invisible to every SIGINT asset in the inventory.
ACOUSTIC
Propellers move air. Air carries sound. The one signature the threat cannot engineer away — and the one signature Ziz is built to hear.
§05 — ENGAGEMENT

Detect. Classify.
Locate.

Passive.
Distributed.
Uncopyable.