Parcel Perform has fine-tuned Amazon Nova models to improve data extraction accuracy from emails, achieving 94.77% accuracy. The optimization reduced inference latency and costs, enhancing e-commerce logistics operations.
Parcel Perform, an AI delivery experience platform, processes millions of emails daily and faced challenges extracting structured data from diverse email formats. Issues included model hallucinations, confusion between similar data types, and high costs of processing HTML emails.
To address these issues, Parcel Perform collaborated with the AWS Generative AI Innovation Center. The aim was to fine-tune Amazon Nova models using customization techniques and parameter optimization to enhance model performance.
The fine-tuned Nova Micro models achieved up to 94.77% extraction accuracy, surpassing the baseline by 16.6 percentage points. Additionally, inference latency was reduced by over 30%, and operating costs were halved, leading to deployment in production for improved logistics.
The fine-tuning process utilized Amazon SageMaker's supervised fine-tuning (SFT) with Parameter-Efficient Fine-Tuning (PEFT). This method enabled customization for specialized entity extraction from e-commerce-related emails, optimizing accuracy and processing efficiency.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
Parcel Perform has fine-tuned Amazon Nova models to improve data extraction accuracy from emails, achieving 94.77% accuracy. The optimization reduced inference latency and costs, enhancing e-commerce logistics operations.