Multinational Consumer Electronics

Designed an LLM-powered workflow for the customer team to analyze user behaviour

Challenge

• Convert natural-language questions into SQL code, run it on the database, and return results to user
• Ensure accuracy and efficiency in processing complex queries while maintaining data privacy and security
• Provide training and support to the analytics team to effectively use the chatbot and interpret the results.

Solution

• Multi-step workflow breaking the problem into smaller tasks
• Dynamic selection of relevant examples shown to the LLM using a Vector database
• Fine-tuning the LLM on specific use cases
• Accuracy uplift through prompt engineering

Business Impact

• Democratizing data access & analytics to non-technical professionals
• Improved understanding of state-of-the-art approaches to LLM text-to-SQL task
• Improved ability to assess challenges and feasibility of reproducing the project for other text-to-SQL use cases
• Potential for increased efficiency and accuracy in generating SQL code from natural language queries