DELBY Intelligence Platform

Forensic Search

Search 4TB of driving data with natural language

crosswalk intersection construction normal driving

Current Clip

Clip ID --

Session --

Timestamp --

Duration --

Loading analysis...

Low Risk

Loading counterfactual analysis...

Select Clip

Detections LiDAR Sync All

Camera Feed + Object Detection

3D LiDAR Point Cloud

0 points

IMU

Ax: 0.00 Ay: 0.00 Az: 0.00

GPS

Lat: 0.0000 Lon: 0.0000

Odometry

Speed: 0.0 km/h Yaw: 0.0°

Detections

🚗 0 🚶 0 🚴 0

Select Session

FRONT

REAR

LEFT

RIGHT

Sensor Timeline 00:00.000

IMU

LiDAR

Odom

Accel X 0.00 m/s²

Angular Z 0.00 rad/s

Speed 0.0 km/h

Text-Conditioned Action Prediction

U-Net Transformer with Gemini Cross-Attention

Enter Driving Instruction

crosswalk intersection stop at light accelerate

The CORTEX model predicts driver actions conditioned on natural language instructions.

Encoder 1 64 channels

Encoder 2 128 channels

Encoder 3 256 channels

Encoder 4 512 channels

Text Encoder Gemini Embedding

Transformer Cross-Attention

Decoder 4 512 channels

Decoder 3 256 channels

Decoder 2 128 channels

Decoder 1 64 channels

Action Output steering, throttle, brake

Predicted Actions

Steering

0.02

Throttle

0.15

Brake

0.72

Hyderabad Coverage

High Risk Zone Medium Risk Low Risk Ambulance

10 Ambulances

94% Coverage

8.2 Avg Response (min)

                            -23%
                            vs Baseline
                        

AMB-001 Jubilee Hills 3 hotspots

AMB-002 Hitech City 2 hotspots

AMB-003 Gachibowli 4 hotspots