Automating Content Quality Control (QC) at Noon

Project Lead – Content & Process Optimization

Challenges:
The key challenges included managing a significant six-month backlog in content quality control (QC) with a monthly ingestion of over 1M new products. The existing manual QC process was time-consuming, costly (~$50,000/month), and led to inconsistencies in content quality. Additionally, the marketplace faced broader issues such as counterfeit products and incorrect listings, which affected customer experience and seller ratings.

Approach:
To address the backlog, I developed a prioritization system focusing on new sellers and top contributors to GMV. Once the backlog was reduced, I identified inefficiencies in the manual QC process and collaborated with the engineering team to explore automation options. We tested multiple Large Language Models (LLMs) such as GPT-4 and Claude 3, eventually selecting Claude 3 due to its 80-85% accuracy. I also refined content guidelines to ensure consistency and built specific use cases to train the LLM for optimal results.

Results:

Reduced QC costs by 50%.
Processed up to 200,000 products per day with 90% of QC volume handled by Claude 3.
Decreased product go-live time from 6 weeks to less than 48 hours.
Enhanced customer experience and improved seller ratings through better content quality control.