Automating Quality Control with Gen AI for Faster, Scalable, and Cost-Effective Image Analysis

Customer Overview

A US-based product photography giant provided an application for auto dealers to take professional-grade photos of car interiors and exteriors using 37 image compositions. Our client polished these images and delivered 4K high-definition outcomes. They provided comprehensive guidelines and training to auto dealers’ personnel to ensure they took photos suitable for editing and polishing.

Project Overview

Using sampling techniques, our client conducted manual Quality Control (QC) of photos coming from auto dealers once a week. Two QC representatives flagged errors, graded quality, and relayed issues to the dealers. As not all images underwent QC, editors had to pause and mark errors on photos, increasing labor costs and processing time. Our client wanted an automated QC solution to analyze every image and improve the feedback loop.

Challenges

Developing an automated, fast, cost-effective QC solution (under time constraints) to analyze every image and send feedback notifications to dealers or stakeholders in the required format.

• The solution must analyze all incoming images from all auto dealers, rather than just a sample, to ensure 100% quality assurance.
• It must assess every important and diverse aspect or quality criterion for each of the 37 image compositions.
• The solution must be capable of analyzing multiple images concurrently to speed up the QC process.
• It must deliver prompt image analysis results in our client’s required format and improve the feedback loop with dealers.
• Given that our client processes approximately two million images monthly, keeping costs low, even with high volumes, is a top priority.
• We must build a workable solution that meets our client’s expectations within a short timeframe and develop a mechanism to enhance the solution’s accuracy.

Solution

Using OpenAI’s GPT-4o multimodal Gen AI model, AWS Lambda, and both frontend & backend development, we built a fast, scalable, cost-effective, automated QC solution.

• To ensure every image is thoroughly analyzed, we defined and established multiple quality criteria for each of the 37 image compositions.
• We used OpenAI’s GPT-4o multimodal model and created prompts that give Pass/Fail results (with the reason) for each quality criterion in every image.
• We speeded up and scaled automated QC by using AWS Lambda to process 10k images concurrently and OpenAI’s plan capable of 10k calls per minute.
• We combined images in a group and used OpenAI’s Batch API, rather than processing them separately, to reduce costs by 50%.
• Integrating this solution with our client’s existing workflow ensured automated QC for every image. Using a customized UI, we generated reports showing pass/fail for each criterion and accepted/rejected photos.
• Using an evaluation framework and human-in-loop techniques, we avoided false positives and improved feedback accuracy.

Benefits

• Replacing the slow, incomplete, labor-dependent QC process with an automated QC solution saved time and costs.
• The dealers’ feedback loop became faster, more efficient and more detailed, improving turnaround time and image quality.
• Dealers received feedback for all images within a day, instead of waiting a week for a sample batch, leading to greater satisfaction.
• Our client strengthened their USP of high-quality outcomes by enhancing QC, boosting brand image and customer trust.

Technology

• GPT-4o
• AWS Lambda
• React
• Node
• Batch API

Industry

• Automobile/Automotive