Refactor hospital suggestion logic in agent executor to ensure location is validated before attempting to find the best matching hospital. Enhance logging for user requests without location and update hospital-related keywords for improved recognition.
Remove GeminiLLM integration from ReportGenerationService, simplifying the report generation process by eliminating the markdown report generation logic. This change focuses on streamlining the service and enhancing maintainability.
Refactor report generation to focus solely on medical data. Introduce MedicalReport interface, streamlining report content by removing unnecessary technical details. Consolidate endpoints to return only JSON format for reports, enhancing clarity and usability. Update report generation logic to align with new structure, ensuring compatibility with existing ComprehensiveReport interface for backward compatibility.
Add integration guide for Report Generation API, detailing endpoints for generating full, markdown, and JSON reports. Update agent executor and triage routes to track tool executions and guidelines count for report generation, enhancing logging for better traceability. Ensure all tool executions are logged for report generation success.
Add Triage Report Implementation Plan with AI Decision-Making features. Introduce new API endpoint for generating complete triage reports, modify agent executor for report generation, and enhance frontend components for user-triggered report generation. Update type definitions to support new report structure and integrate location services for nearby medical facilities. Include detailed verification plan and implementation timeline.
Add API Integration Guide for Triage Health Check endpoint with detailed request and response specifications. Update agent executor to enforce concise and clear response formatting guidelines in Vietnamese, ensuring compliance with medical guidelines and improving user interaction clarity.
Refactor intent classification logic in IntentClassifierService to prioritize image presence and improve casual greeting detection. Introduce a structured priority system for intent classification, enhancing accuracy for triage and disease information queries while ensuring casual greetings are only identified when appropriate. Update the isCasualGreeting method to exclude medical terms from greeting detection.
Implement intent classification for casual greetings in the IntentClassifierService and update MedagenAgent to route user queries based on classified intent. Enhance response handling in triage routes to utilize markdown messages when available. Extend TriageResult type to include a message field for markdown responses from the LLM.
Add detailed logging for prompts and input data in agent executor to improve debugging and traceability. This includes user text, guidelines count, and conversation context before sending requests to the LLM.
Update educational response prompts in agent executor to enhance clarity and ensure compliance with Vietnamese guidelines. Adjusted language for user questions and response structure, emphasizing the requirement for responses to be entirely in Vietnamese and providing specific instructions for JSON response formatting.
Enhance CV result handling in agent executor by filtering based on confidence threshold and updating logging. Adjust match threshold and count in RAG service for improved guideline retrieval.
