ANOMGRID_
Building in stealth · v0.1
Reach out →
SOMETHING AWESOME COMING FOR THE ROADS THAT BREAK MODELS FOR THE WORLD THAT REFUSES TO BEHAVE STAY TUNED SOMETHING AWESOME COMING FOR THE ROADS THAT BREAK MODELS FOR THE WORLD THAT REFUSES TO BEHAVE STAY TUNED
// PROJECTANANT_01
// STATUSPROTOTYPE
// REACHGLOBAL
// DOMAINEMBODIED_AI

Built for the moment
AI meets the unknown.

Models break where reality refuses to behave. AnomGrid works in that gap — the layer beneath every embodied AI for the moments their training never saw.

Meet Anant → See the problem
FIELD_FEED · DASHCAM_07
[ANANT_01.PROTO]
Autonomous vehicle in unstructured Indian traffic
BUS · 0.92 TRUCK · 0.88 EGO · TRACKED RICKSHAW · 0.85 MOTORCYCLE · 0.81 PEDESTRIAN · CLOSE UNCLASSIFIED · ? ⚠ MODEL CONFIDENCE: 0.41
UNSTRUCTURED VIDEO + AUDIO 5 LAYERS · 4 CLASSIFIED
// LABEL_PASS_03
FIELD DATA · COLLECTED 200 HOURS · DASHCAM VIDEO + AUDIO · TIME-SYNCED HORN TYPOLOGY · CAPTURED UNCONTROLLED INTERSECTIONS FIELD DATA · COLLECTED 200 HOURS · DASHCAM VIDEO + AUDIO · TIME-SYNCED HORN TYPOLOGY · CAPTURED UNCONTROLLED INTERSECTIONS
// FIELD_DATA · LIVE_COUNTER
0hr
Dashcam data collected
AND COUNTING · ON THE GROUND
// MODALITIES
0×
VIDEO + AUDIO · frame-synced
// LAYERS
0
5 negotiation layers · 4 classified
// COVERAGE
∞
Horn typologies · contextual signals
01 THE PROBLEM

Every AI is certain until reality
refuses to cooperate.

Self-driving cars trained on Western highways collapse on roads where horns are language and right-of-way is decided by mass and confidence. Robots trained in clean labs break the moment the gas runs out, the conveyor jams, the world refuses the script.

These aren't edge cases. This is how most of the world actually operates. The data nobody collected. The signals nobody labeled. The intelligence layer nobody built.

We work in that gap.

// FIELD_LOG · CHAOS_INDEX LIVE
[01]Lane discipline12%
[02]Signal compliance34%
[03]Surface integrity21%
[04]Negotiation density88%
[05]Foundation-model fit19%
// SAMPLE_WINDOW · 30D// QUALITATIVE · FIELD
FIELD_01 CONF 0.31 Autonomous car in dense Indian traffic
// CASE_01
The road that breaks the map.
FIELD_02 CONF 0.28 Dense scooter and motorcycle traffic
// CASE_02
The signals nobody labeled.
FIELD_03 CONF 0.22 Damaged broken road surface with potholes
// CASE_03
The blind edge of every model.
FIELD_04 CONF 0.34 Dense urban traffic
// CASE_04
Density beyond any benchmark.
ANANT_01FIELD STUDY · UNSTRUCTURED ROADSVIDEO + AUDIO5 LAYERS · 4 CLASSIFIED ANANT_01FIELD STUDY · UNSTRUCTURED ROADSVIDEO + AUDIO5 LAYERS · 4 CLASSIFIED
FLAGSHIP // PROGRAM_01

A n a n t .

Our first study. The roads that break models, and the implicit grammar that decides who proceeds.

Mobility, beyond the western lane.

Anant is our entry into the largest unmapped territory in autonomous mobility — the unstructured road. Not a country. A condition. The condition under which most of the world drives.

We are not solving for one geography. We are solving for everywhere the road refuses to behave — every emerging market, every chaotic intersection, every place where the gap between training and reality is largest.

What we collect, how we model, what makes it work — that part stays quiet until it ships.

200hr
Dashcam data collected · field
2×
Modalities · video + audio sync
5
Negotiation layers · 1 documented · 4 classified
∞
Horn typologies · contextual signals
// MODALITIES · WHAT_WE_CAPTURE VIDEO + AUDIO · TIME-SYNCED
The visual stream

Front-facing dashcam · 1080p · 30 fps · GPS + IMU · weather + time tags · captured across uncontrolled intersections.

The acoustic stream

Stereo cabin + external mic · horn typology, sequence, intensity envelopes · synchronised frame-perfect to the visual track.

At the heart of Anant is a negotiation language.

Five layers. One signal. The grammar of every uncontrolled intersection on Earth.

[01]

Layer 1 — Physical Signals

The observable kinematics: position, velocity, acceleration, trajectory, gap acceptance, deceleration profiles. The "what is happening" layer. Documented incompletely in IDD, KITTI, nuScenes.

documented
[02]

Layer 2 — Acoustic Signals

The sound layer that no AV dataset has ever encoded. Horn typology, duration, sequence, spatial origin, intensity envelopes synchronized to the visual frame. The vocabulary of negotiation in unstructured traffic.

classified
[03]

Layer 3 — Kinematic Intent Signals

The pre-movement micro-signals — sub-50cm creeps, deceleration as deference, lateral angle as claim. The body language of vehicles and pedestrians, captured 1.5–3 seconds before any classified action begins.

classified
[04]

Layer 4 — Social Hierarchy Signals

The implicit precedence structure — vehicle class, role, locality, demographic. Encoded as a graph of right-of-way priors that varies by region and context.

classified
[05]

Layer 5 — Contextual Modifiers

Temporal, environmental, and cultural state that transforms the meaning of every other layer — time-of-day, weather, festival calendar, presence of authority, religious or civic events.

classified
02 THE BIGGER ARC

Anant is the first study. The grammar repeats — wherever a system meets the world without a script, the same problem returns. We work one domain at a time, with patience the foundation-model era has forgotten.

// TODAY: AV // NEXT: ROBOTICS // AFTER: ANYTHING_EMBODIED // HORIZON_05_YEARS
BUILDING IN STEALTH · v0.1

Something awesome
coming...

We are prototyping the layer — quietly, carefully, in the field. Stay tuned.

// PROTOTYPE BUILD42% · IN_PROGRESS
0hr
DASHCAM DATA · COLLECTED
0
SIGNAL LAYERS
0
CLASSIFIED LAYERS
[CONNECT]

Build with the
layer beneath.

GENERAL → FOUNDER →