AI-Powered Management Decision Support System (MDSS) Development Notes3
Today, I focused on improving the chat functionality of my AI-powered Management Decision Support System (MDSS).
One of the main issues in the previous version was that the AI attempted to perform business analysis regardless of the user's input.
For example, even when a user entered "Hello," the system responded with sales and profit analysis, which was unnatural for a chat interface.
The root cause was that the prompt used for normal conversations was also designed to trigger business analysis.
To solve this, I redesigned the prompt so that the system references the current analysis data only for business-related questions, while responding to all other inputs as a normal conversational AI.
As a result, the system now behaves as intended:
"Hello" receives a normal conversational response.
"Update" triggers AI-driven business analysis and stores the results in the database.
"What-if" simulations and follow-up requests (such as "Translate this into Japanese" or "Add another condition") invoke AI reasoning, but the results are displayed only in the chat and are not saved to the database.
This improvement allows the chat interface and the MDSS to coexist naturally within a single application.
On the other hand, What-if analysis (for example, "What happens if sales decrease by 30%?") and follow-up requests requiring additional analysis or translation depend on the reasoning capability of an LLM. Therefore, these features are designed to call the API when necessary.
Determining which tasks should be handled by the AI and which should be managed by the database or the application itself is a key architectural decision.
This separation of responsibilities has a significant impact on both the user experience and the operational cost of the system.
Comments