Submit Your Database

Blueprint for Chatbot Prompt Templates: 1. **Intent Classification Prompt**: - **Purpose**: Classify the user’s query to determine its intent and relevance to the database. - **Structure**: - Input variables: user_query, available_databases_summary, database_content_summary, conversation_summary - Template: - Describe the AI’s role as a specialized chatbot for your database’s domain (e.g., plant ncRNAs or fusion transcripts). - Specify the databases served (e.g., MyDatabase). - Instruct the AI to interpret queries like “what data do you have on [topic]?” as data_preview requests. - List classification categories: METADATA_DIRECT_ANSWER_PREFERRED, METADATA_SQL_FALLBACK_PREFERRED, DATA_RETRIEVAL, AMBIGUOUS, TYPO, GENERAL_CONVERSATION, OUT_OF_SCOPE. - Define each category with examples relevant to your database’s content. - Output only the classification string. - **Example Usage**: “Classify whether a query like ‘List fusion transcripts in rice’ is a data retrieval request or if ‘What is the time?’ is out of scope.” 2. **SQL Planning Prompt**: - **Purpose**: Generate SQLite-compatible SQL queries or metadata responses based on the query’s intent. - **Structure**: - Input variables: user_query, classified_intent, db_file_name, schemas_json, knowledge_graph, database_table_mapping_info, DISPLAY_ROW_LIMIT, retrieved_knowledge_context, conversation_summary, previous_query_summary - Template: - Define the AI’s role as a specialized chatbot for your database. - Specify that output must be a JSON object wrapped in and . - Provide database overview (e.g., key columns like gene1, gene2, species). - Outline rules for handling query types (e.g., gene ID queries check multiple tables, expression queries target specific columns). - Include SQL best practices (e.g., quote table/column names, use single quotes for strings). - Define JSON structure: query_type, analysis_plan, direct_answer_from_metadata, queries (with sql, target_table, database_conceptual_name, description, purpose_type, display_columns_hint). - **Example Usage**: “Generate SQL to retrieve fusion transcripts for gene AT1G12345 from MyDatabase.” 3. **Summary Interpretation Prompt**: - **Purpose**: Rephrase system-generated data summaries into user-friendly, conversational responses. - **Structure**: - Input variables: user_query, analysis_plan_from_stage1, textual_data_summary_from_python, knowledge_graph_snippet, conversation_summary, query_type_from_stage1, DISPLAY_ROW_LIMIT - Template: - Specify the AI’s role and database focus. - Instruct to output a JSON object with “summary” and “databases_conceptually_involved” fields. - Require the summary to be based solely on the system-generated data summary, using conceptual database names. - Define rules for handling data previews, aggregations, errors, and common item analyses (e.g., normalize species like ‘oryza sativa japonica’ to ‘oryza sativa’). - Emphasize professional tone, replacing underscores with spaces for readability. - **Example Usage**: “Rephrase a system summary stating ‘Found 150 fusion transcripts’ into: ‘MyDatabase identified 150 fusion transcripts for your query.’” **Instructions**: - Use this blueprint to create prompts tailored to your database’s schema and content. - Ensure prompts handle your database’s specific data types (e.g., lncRNAs, fusion transcripts) and unsupported queries (e.g., siRNAs). - Test prompts with your sample queries to verify accurate intent classification and data retrieval.