Collision AI

Inspiration

What it does

How we built it

Challenges we ran into

Accomplishments that we're proud of

What we learned

Team name: Collision AI

The problem to solve: The problem to solve: Halo is a collision repair group that has a unique process for repairing vehicles. In most shops, you start with a full estimate and agree on a price with the insurer. We've found that this actually just adds an unnecessary delay and layer of complexity, so instead, we do the estimate and agreeing after the car has been repaired. Since we have a fixed price model too, we don’t have to worry about making the repair profitable through scummy estimating or trying to get stuff added. We just try to repair the car in a safe manner as quickly as possible. Usually, you triage the cars coming into a shop when you do the estimating up-front, but as previously stated, this causes numerous other problems. Instead, we need to find a way to effectively triage the cars without involving a whole estimating team at the start of the process. Currently, at Halo, the triaging is done by the claims team, who have no technical training, and we want them to focus on providing an extraordinary customer experience, not creating pseudo-estimates at the start of a claim. We end up with total losses on-site, hub-sized jobs at spokes, and hubs filled with easy work they can’t complete quickly. We’ve been trying to figure out how to solve this issue, but at the moment the only explicit fields we have to go on are the “Drivable or Non-drivable” status from the insurance FNOL. Damage descriptions alone aren't good enough either since neither the vehicle owner nor the insurance claim handler usually know enough about repairing vehicles to provide an accurate representation of the actual damage. We end up with “oh it's just a scratch” jobs that end up being the full side caved in, and “my baby is demolished” ones that are just a scratched bumper cover.

The Solution: Examines images and information available for a vehicle/claim, explains what it sees, and lays out its logical thinking for determining if the vehicle is a likely total loss, repairable by a spoke site, or must go to a hub location. Also determines drivability of the vehicle.

Built With

• openai
• streamlit