Ask-a-Metric is an innovative WhatsApp-based AI data analyst that uses Large Language Models (LLMs) to answer SQL database queries, streamlining data access for decision-making in the development sector. Initially facing challenges with accuracy and scalability, the team tested an agentic approach using CrewAI, resulting in improved accuracy but high costs and slow response times. Combining learnings from the agentic experiments, they developed a pseudo-agent pipeline that combines the speed and cost efficiency of the original simple version with the accuracy of the agentic approach. This new pipeline optimally balances response time, cost, and accuracy, paving the way for further enhancements and features in the future.
https://idinsight.github.io/tech-blog/blog/aam_pseudo_agent/