Inspiration
Traditional business analytics for local life and retail industries has long been inefficient and unreliable. Business teams cannot write SQL and rely heavily on data analysts for basic queries. Key metrics including GMV, repurchase rate and verification rate lack unified statistical standards. Raw LLM-generated SQL also causes serious issues such as unauthorized data access, slow queries and calculation errors. In addition, manual multi-dimensional attribution analysis is time-consuming and leads to excessive waste of database and model resources. These pain points drive us to build an intelligent, secure and automated business analytics Agent platform for multi-tenant scenarios.
What it does
This platform converts natural language business questions into standardized, verifiable data analysis results. It covers four core business scenarios with a full closed-loop analysis system: basic metric query, multi-dimensional trend comparison, business anomaly attribution, and data-driven business decision support. The system outputs accurate metrics, visual charts, detailed attribution results and actionable operational suggestions with complete data basis, helping teams analyze business data without professional SQL skills.
How we built it
We constructed a complete automated analysis workflow: natural language input → business metric and intent understanding → semantic schema recall → intelligent analysis planning → safe SQL generation → permission and cost verification → query execution → multi-dimensional attribution mining → chart and business report output. Based on structured semantic models and permission control mechanisms, we realized standardized, secure and low-consumption intelligent data analysis for multi-tenant business scenarios.
Challenges we ran into
First, it was difficult to guarantee the safety and accuracy of LLM-generated SQL. We solved unauthorized access, slow query and statistical error problems by building strict verification rules. Second, inconsistent business metric calibers led to confusing analysis results, which we fixed with unified semantic metric standards. Third, repeated queries caused huge resource consumption, and we adopted parallel planned analysis to optimize database and model resource utilization.
Accomplishments that we're proud of
We successfully built a usable, secure and efficient AI business analysis Agent that eliminates reliance on technical SQL capabilities for business staff. The platform unifies industry metric standards, realizes automatic full-link attribution analysis for business anomalies, and effectively reduces database and model resource consumption. It can independently complete from data query, trend comparison to decision suggestion output, forming a complete industrial-level business analysis closed loop.
What we learned
We learned how to apply large language models to real industrial business scenarios rather than simple demo scenarios. We mastered the method of balancing AI automation efficiency with business data rigor and security. We also accumulated practical experience in business metric standardization, LLM SQL security optimization, multi-dimensional data attribution analysis and intelligent resource scheduling for enterprise-level data systems.
What's next for Merchant Insight
We will further enrich business scenario adaptation capabilities to cover more retail and local life service business models. We plan to optimize the intelligent prediction module to realize pre-judgment of business trends and user churn risks. Meanwhile, we will strengthen multi-tenant personalized configuration functions and improve the intelligent iteration ability of analysis strategies to provide more refined, intelligent and customized business analysis services.
Log in or sign up for Devpost to join the conversation.