// AI-Adaptive Vision Headlights

See everything.
Blind
no one.

HALO is a real-time adaptive headlight system that perceives the road in three dimensions and selectively shapes its beam — illuminating the path ahead while masking glare around oncoming drivers, pedestrians, and cyclists.

Live Beam Simulation · Stereo Depth + Selective Masking
Adaptive Beam ShapingStereo Depth LocalisationEdge Inference at the SourceGlare-Critical Region Masking Adaptive Beam ShapingStereo Depth LocalisationEdge Inference at the SourceGlare-Critical Region Masking
01 — Company Vision

Light should think before it shines.

Trismegistus Works exists to rebuild the most fundamental piece of automotive safety — the headlight — for an age where vehicles can finally see. For over a century, headlights have been blunt instruments: bright or dim, on or off, oblivious to the world they illuminate.

HALO is our answer. A perception-first lighting system that fuses stereo vision, depth-aware localisation, and on-edge intelligence to produce light that is precise, adaptive, and considerate. We do not switch between high-beam and low-beam. We sculpt the beam, frame by frame, across sixteen independently controlled segments — illuminating the road while preserving the vision of every driver, pedestrian, and cyclist sharing it.

Our mission is to make night-time driving demonstrably safer for two-wheelers first, then four-wheelers, then everything that moves after dark. We are building toward a future where adaptive perception is not a luxury feature, but the default expectation of every vehicle on the road.

// 01
Perceive, then illuminate
Every photon emitted is the product of a decision. Stereo cameras and a learned perception model identify glare-critical regions before the beam is shaped — not after.
// 02
Hybrid by design
Data-driven learning handles perception. Deterministic control logic handles actuation. The result is a system that is robust, explainable, and ready for the safety constraints of automotive deployment.
// 03
Built at the edge
All inference runs on-vehicle, with no cloud dependency. Sub-frame latency is non-negotiable. The architecture scales from a Jetson prototype to a custom FPGA built for production.

A high-beam decides everything.
HALO decides per pixel.

02 — Beam Pattern Comparison
// Static · Adaptive · Selective
// CONVENTIONAL — Binary Switching
Static Beam Pattern
Either floods the road and blinds the oncoming driver, or dims globally and sacrifices forward visibility. No spatial awareness. No middle ground.
// HALO — Selective Masking
Adaptive Matrix Beam
Sixteen LED segments are individually attenuated only where glare would occur. Maximum illumination is preserved everywhere else, frame after frame.
03 — System Architecture

Four subsystems, one continuous loop from photon to photon.

// 01 · SENSING
Stereo Camera Pair
Two synchronised cameras capture the forward field of view at 30 FPS. Calibrated baseline enables disparity-based depth recovery for every detected entity.
04 — The Beam, In Pieces

Sixteen pixels of intent.

The HALO matrix array divides the forward beam into sixteen Cree XP-G3 LED segments, each individually addressable through an EVQ7228 / MPQ7228 driver bank. Every frame, the perception pipeline emits an attenuation map: which segments stay bright, which dim, which fully extinguish.

Move your cursor across the array. Each segment responds independently — the same logic that, on the road, decides whether to illuminate a tree or shield an oncoming rider's eyes.