187DEF

    REAP-T2 "Scythe" is live: zero real frames in training, 88% on real footage. Request the brief →

    EU-first defense engineering / Bucharest, Romania

    Europe's Defense AI Foundry.

    Request a brief →Talk to us
    01COMPANY

    We build military detection, segmentation & tracking models and the synthetic data factory training them.

    02CAPABILITIES
    C.01

    PERCEPTION

    Object detection, segmentation, and tracking for aerial threats, measured on real footage. Detection is trained entirely on synthetic imagery; the track layer stitches its sparse detections into near-continuous, engagement-grade tracks across thermal and daylight sensors.

    Delta-wing aerial threat detected in thermal imagery over forest, detection box locked

    MEASURED ON REAL OPERATIONAL FOOTAGE

    88%
    Center-hit rate

    Detection on real thermal intercept footage of Shahed-136s.

    90–100%
    Track coverage

    Frames with a confirmed track, on real operational footage.

    370–530ms
    First-lock latency

    Well under a second across sensors.

    0
    Real frames in training

    Detection trained on synthetic imagery only, validated on real intercept footage.

    2 systems

    C.02

    SYNTHETIC DATA

    Generate labeled multi-sensor imagery of air and ground targets, with no flight or range clearances, field crews, or sensor assets.

    A simulated battlefield engine with high-fidelity, physics-based sensor models: labeled visible, IR, and low-light imagery of aerial threats and ground vehicles over real-world and procedural terrain. Labels are pixel-accurate at generation time, and the output is graded by what it trains: detection performance on real footage.

    Synthetic thermal scene: a main battle tank in forest, segmented and boxed, rendered by RESYNTH

    GRADED DOWNSTREAM / DETECTOR TRAINED ON SYNTHETIC ONLY

    88%
    Downstream detection

    Reached on real thermal footage by a detector trained on synthetic imagery alone.

    0
    Real capture required

    No flight or range clearances, field crews, or sensor assets.

    100+
    Models in library

    Aerial threats and ground vehicles, each built from reference material.

    2days
    New model turnaround

    From reference material to an added entry in the library.

    1 system

    03RESEARCH
    R.01

    Multi-sensor fusion

    Cross-modal correlation between Visible (RGB) and Thermal (LWIR) sensor streams.

    R.02

    Edge inference

    Quantized detection pipelines for SWaP-constrained platforms.

    Research programs →
    04CONTACT

    Specs, datasets, partnerships, and research.

    Correspondence
    Program inquiries

    Submit a brief for evaluation.

    Submit a brief →