Improving User Experience for Exploring Counters and Insights Data: A Data Science Approach

In this guest post co-written by Ori Nakar from Imperva, we explore how Imperva Cloud WAF protects websites against cyber threats and blocks billions of security events daily. With the goal of improving user experience in exploring counters and insights data, Imperva utilized a large language model (LLM) to enable natural language search queries for their internal users.

The challenge of ensuring quality in constructing SQL queries from natural language was addressed using a data science approach. By creating a static test database, a test set with known answers, and examples for translating questions to SQL, Imperva fine-tuned their LLM-based application.

Amazon Bedrock, a managed service offering high-performing foundation models, was instrumental in the experimentation and deployment process. With the ability to easily switch between models and embeddings, Imperva was able to improve accuracy and optimize costs for their application.

The key takeaway from this project is the importance of creating a test set with measurable results to track progress and compare experiments. By leveraging Amazon Bedrock and following a data science approach, Imperva was able to successfully construct SQL queries from natural language and enhance the user experience for their application.

If you’re interested in experimenting with natural language to SQL, check out the code samples in the GitHub repository mentioned in the post. This workshop provides modules that build on techniques for solving similar problems using LLM-based applications.

Overall, this collaboration between Imperva and Amazon Web Services showcases how innovative solutions can enhance data accessibility and user experience in the cybersecurity space. With the right tools and approaches, organizations can streamline processes and improve outcomes in handling security threats.