IRIN LABS PVT LTD / EST. 2026

The vision pipeline
for complex edge
cases.

IRIN Labs Pvt Ltd provides infrastructure-grade datasets and active inference for rare, safety-critical moments on complex road networks. Built on L4-grade autonomous principles, our pipeline enables predictive safety, even where standard datasets fail. Deploying across NH48 and key urban routes.

GET EARLY ACCESS EXPLORE THE NETWORK
BUILT FOR INDIAN ROAD CONDITIONSEDGE-FIRST BY DESIGN
EDGE CAPTURE / 001 ON DEVICE
NH 48 / KA / IN HDR · 120 FPS
MODEL EVENT / EXAMPLE FEED LIVE DATA STREAM SIMULATED
EVENT 01ANIMAL_CROSSING // 0.9817 m · 14:32:01 · E-0473
EVENT 02OVERLOADED_2W // 0.9111 m · 14:32:01 · E-0474
01 A CONTINUOUS DATA PIPELINE
ON-DEVICE FILTERING×RARE EVENT CAPTURE×SEARCH-READY DATA×INDIAN ROAD CONTEXT
01 — DATASET INFRASTRUCTURE

Built for the roads
the world missed.

IRIN turns real-world fleet miles into structured, search-indexed edge-case data for ADAS and autonomous systems operating in Indian road conditions.

EDGE CAPTURE RATE120 FPS

Dual 1080p HDR streams evaluated on-device.

INFERENCE LATENCY< 40ms

On-device rare obstacle classification target.

MODEL PROFILE8-bit quantized

Qwen-VL on low-power Jetson and Raspberry Pi 5 edge nodes.

REFERENCE CONFIGURATION / 001

Structured at the source.

Example event record and sensor envelope for a two-camera vehicle unit.

CAMERA INPUTS
2 × 1080p · HDR · up to 120 FPS
EDGE BUFFER
30s circular RAM · in-memory inference
EVENT CLIP
10s · 5s pre / 5s post trigger
DETECTOR CLASSES
Obstacles · vehicles · trajectory · environment
GET /api/v1/telemetry/query?hazard_type=UNSTRUCTURED_OBSTACLE&min_confidence=0.90JSON
{
  "status": "200_OK", "query_execution_ms": 11,
  "total_records": 1842,
  "results": [{
    "clip_id": "clip_nh48_9942",
    "timestamp": "2026-10-08T02:14:00Z",
    "location": { "lat": 12.9716, "lng": 77.5946, "route": "NH-48" },
    "classification": "ANIMAL_INCURSION", "confidence": 0.94,
    "s3_video_uri": "s3://irin-data-store/clips/clip_nh48_9942.mp4",
    "annotation_uri": "s3://irin-data-store/labels/clip_nh48_9942.json"
  }]
}
02 — TECHNICAL ARCHITECTURE

Only the signal
leaves the vehicle.

A privacy-aware edge pipeline cuts routine footage before it reaches the network. Rare events are structured locally, then synchronized when connectivity allows.

98.4%LESS RAW DATA SENT TO CLOUDSAFETY TAXONOMY FOR L2+–L4 VALIDATION
STAGE 01 / PIPELINE DETAIL

Forward and side cameras feed a low-power NVIDIA Jetson or Raspberry Pi 5. A 30-second RAM buffer supplies recent frames to quantized Qwen-VL inference.

↗ENCRYPTED STORE-AND-FORWARD
Sync on cellular availability READY
03 — EARLY ACCESS

Build with the
road ahead.

For fleet operators and autonomy teams building for Indian roads.

DIRECT INQUIRIEShello@irinlabs.in
REQUEST EARLY ACCESS* REQUIRED
Direct response from our team.