Our Inspiration: The Secret of DAS
We're all F1 fans, and we were fascinated by the story of Mercedes "Dual-Axis Steering" (DAS) in 2020. Onboard footage showed Lewis Hamilton pushing and pulling his steering wheel on the straights. It wasn't a malfunction—it was a secret weapon. He was actively adjusting the toe angle of the front wheels, a brilliant and subtle strategic move.
This got us thinking: How would a competitor even detect this?
A normal "spot the difference" tool is useless here. It would compare two frames and flag everything the vibration, the shadows, the reflections. It would be 100% "noise." This is the problem that inspired Spotter. In high-stakes environments like F1 or manufacturing, the challenge isn't just finding "change." It's finding the one critical, meaningful change (the "signal") hidden in a sea of visual noise.
What We Plan to Build: An Analyst, Not a Calculator Current tools are just 'calculators.' They subtract Image A from Image B and give you a messy heatmap. This creates "alert fatigue", when everything is flagged, real emergencies are missed.
Spotter will be different. It's not a calculator; it's an AI analyst. We plan to build an agentic AI workflow that uses a chain of tools to replicate human expert reasoning. It won't just find pixels; it will find insight.
Our Plan: How Spotter Will Work When our agent is given two images or videos (like the DAS-enabled car), it will autonomously create and execute a plan:
Align: First, it will use OpenCV to find key features and perfectly align the two images or videos, correcting for camera shake and vibration.
Identify: It will then run a YOLO (or similar) object detection model to find the parts it understands, like front wheel, chassis, and steering wheel.
Analyze: This is the core. It will run specialized tools. For the DAS example, it would run a pose estimator on the steering wheel (to detect the 'push' vs. a 'turn') and a geometric analyzer on the front wheel (to measure the angle).
Reason: Finally, a Large Language Model (LLM) will synthesize these findings. It will be trained to connect the "push" and the "angle change," while actively discarding the lighting changes as "noise."
Report: The final output won't be a messy image. It will be a clean report:
NOISE (Ignored): Reflection on chassis.
CRITICAL (Flagged): DAS System Detected. Toe angle changed by 0.7 degrees following non-rotational steering input.
