AI That Delivers: How VirtueCloud Helped a Manufacturer Transform with AWS GenAI

In collaboration with AWS, VirtueCloud has successfully implemented a Generative AI solution that dramatically improved the workflow of manufacturer. By integrating advanced AI technologies, we achieved a remarkable 94% reduction in order-preparation time, transforming a traditionally cumbersome process into a streamlined operation. This case study highlights the challenges faced, the innovative solutions deployed, and the impressive results achieved.

The Challenge

The client faced significant challenges with their existing workflow. Client specifications arrived in various formats, including inconsistent PDF diagrams and tables. Support engineers were tasked with manually rebuilding Bills of Materials (BOMs), checking stock levels, querying vendors, and opening jobs. This manual process resulted in a high error rate, a lack of real-time insights, and prolonged procurement cycles.

Our GenAI Solution (AWS Tech Stack)

To address these challenges, we implemented a comprehensive Generative AI solution utilizing the AWS tech stack. Below is a breakdown of the layers involved:

Optimisations & Lessons

Throughout the implementation, we identified several areas for optimization:

• PDF Variance: We added Bedrock post-processing to enhance Textract's accuracy by 22%.
• Prompt Drift: We iterated on few-shot prompts, successfully reducing hallucinations to below 1%.
• API Rate-Limits: We batched Lambda calls and cached vendor quotes in S3, achieving a 10× reduction in latency.
• Cost Control: We transitioned to SageMaker serverless and Triton, resulting in 40% cheaper inference costs.

Results

The implementation of our GenAI solution yielded impressive results:

• 94% faster BOM-to-PO cycle: Reduced from 4 hours to just 15 minutes.
• Zero manual data entry: Achieved an error rate of less than 0.5%.
• 40+ engineer hours saved per month: Significantly increased operational efficiency.
• Real-time pricing: Enabled better margin decisions for the client.