Mobile Tech RX (MTRX) is transforming the auto reconditioning industry by delivering an auto repair app built by technicians, for technicians. The Mobile Tech RX app makes it easy for technicians to estimate, invoice, manage teams, and capture data on-the-go and from a phone.
Mobile Tech RX came to KUNGFU.AI with some lofty goals. They wanted to build an automated pricing tool so customers could get a quote to repair their car's bodywork in seconds. The MTRX team had undertaken a large internal labeling effort, manually labeling over three thousand images of dents in vehicles, across all shapes, sizes, models, colors and make. They then annotated and denoted the pictures with the labels “dent”, “hail”, “paint”, or “conventional” (beyond repair).
As the KUNGFU.AI team dove in, they ended up discovering MTRX’s problem. Although it was well-intentioned, MTRX’s labeling system was not ideal. MTRX shared with us their initial attempt at the problem set, which was an out-of-the-box object detection model. This model underperformed and did not give them the ability to price individual dents reliably.
KUNGFU.AI researched this problem set, and identified a better model that had the capability to produce better object detection results, and to produce a per-dent pricing output. This new, better model choice offered higher accuracy, better configurability and tuning, and enabled MTRX to add additional classification or regression outputs per dent when MTRX decided to increase their model’s capability.
Additionally, during the project, to avoid an expensive and redundant labeling effort, the KUNGFU.AI team leveraged classic data science approaches to make use of the existing deficient data set to achieve a useful model.
We at KUNGFU.AI pride ourselves on leaving our clients with a working solution, and the knowledge to operate, and even extend the solution; during our engagement we had weekly sessions with the MTRX technical team to teach them Machine Learning and Deep Learning fundamentals, and help them understand every component of our delivered solution.
Mobile Tech RX (MTRX) is transforming the auto reconditioning industry by delivering an auto repair app built by technicians, for technicians. The Mobile Tech RX app makes it easy for technicians to estimate, invoice, manage teams, and capture data on-the-go and from a phone.
Mobile Tech RX came to KUNGFU.AI with some lofty goals. They wanted to build an automated pricing tool so customers could get a quote to repair their car's bodywork in seconds. The MTRX team had undertaken a large internal labeling effort, manually labeling over three thousand images of dents in vehicles, across all shapes, sizes, models, colors and make. They then annotated and denoted the pictures with the labels “dent”, “hail”, “paint”, or “conventional” (beyond repair).
As the KUNGFU.AI team dove in, they ended up discovering MTRX’s problem. Although it was well-intentioned, MTRX’s labeling system was not ideal. MTRX shared with us their initial attempt at the problem set, which was an out-of-the-box object detection model. This model underperformed and did not give them the ability to price individual dents reliably.
KUNGFU.AI researched this problem set, and identified a better model that had the capability to produce better object detection results, and to produce a per-dent pricing output. This new, better model choice offered higher accuracy, better configurability and tuning, and enabled MTRX to add additional classification or regression outputs per dent when MTRX decided to increase their model’s capability.
Additionally, during the project, to avoid an expensive and redundant labeling effort, the KUNGFU.AI team leveraged classic data science approaches to make use of the existing deficient data set to achieve a useful model.
We at KUNGFU.AI pride ourselves on leaving our clients with a working solution, and the knowledge to operate, and even extend the solution; during our engagement we had weekly sessions with the MTRX technical team to teach them Machine Learning and Deep Learning fundamentals, and help them understand every component of our delivered solution.
