Update modules/orchestrator.py

modules/orchestrator.py

@@ -1,158 +1,108 @@
-# modules/orchestrator.py
+# modules/prompts.py
 """
-The Central Nervous System of Project Asclepius.
-This module is the master conductor, orchestrating high-performance, asynchronous
-workflows for each of the application's features. It intelligently sequences
-calls to API clients and the Gemini handler to transform user queries into
-comprehensive, synthesized reports. (v1.1)
+Central repository for all Gemini prompt engineering.
+This is the "soul" of the AI, defining its persona, tasks, and output structure.
+(v1.2 - The "Insight Engine" Upgrade)
 """
 
-import asyncio
-import aiohttp
-from itertools import chain
-from PIL import Image
-
-# Import all our specialized tools
-from . import gemini_handler, prompts, utils
-from api_clients import (
-    pubmed_client,
-    clinicaltrials_client,
-    openfda_client,
-    rxnorm_client
+# The non-negotiable disclaimer that precedes every major output.
+DISCLAIMER = (
+    "**⚠️ IMPORTANT DISCLAIMER: This is an AI-powered informational tool and NOT a substitute for professional medical advice.** "
+    "The information provided is for educational and research purposes only. "
+    "It is generated by synthesizing publicly available data and may contain inaccuracies or be incomplete. "
+    "**ALWAYS consult a qualified healthcare professional for diagnosis, treatment, or any medical concerns.** "
+    "Never disregard professional medical advice because of something you have read here."
 )
 
-# --- Internal Helper for Data Formatting ---
-# (This helper function remains unchanged)
-def _format_api_data_for_prompt(api_results: dict) -> dict[str, str]:
-    formatted_strings = {}
-    pubmed_data = api_results.get('pubmed', [])
-    if isinstance(pubmed_data, list) and pubmed_data:
-        lines = [f"- Title: {a.get('title', 'N/A')} (Journal: {a.get('journal', 'N/A')}, URL: {a.get('url')})" for a in pubmed_data]
-        formatted_strings['pubmed'] = "\n".join(lines)
-    else:
-        formatted_strings['pubmed'] = "No relevant review articles were found on PubMed for this query."
-    trials_data = api_results.get('trials', [])
-    if isinstance(trials_data, list) and trials_data:
-        lines = [f"- Title: {t.get('title', 'N/A')} (Status: {t.get('status', 'N/A')}, URL: {t.get('url')})" for t in trials_data]
-        formatted_strings['trials'] = "\n".join(lines)
-    else:
-        formatted_strings['trials'] = "No actively recruiting clinical trials were found matching this query."
-    fda_data = api_results.get('openfda', [])
-    if isinstance(fda_data, list):
-        all_events = list(chain.from_iterable(filter(None, fda_data)))
-        if all_events:
-            lines = [f"- {evt['term']} (Reported {evt['count']} times)" for evt in all_events]
-            formatted_strings['openfda'] = "\n".join(lines)
-        else:
-            formatted_strings['openfda'] = "No specific adverse event data was found for this query."
-    else:
-        formatted_strings['openfda'] = "No specific adverse event data was found for this query."
-    vision_data = api_results.get('vision', "")
-    if isinstance(vision_data, str) and vision_data:
-        formatted_strings['vision'] = vision_data
-    elif isinstance(vision_data, Exception):
-        formatted_strings['vision'] = f"An error occurred during image analysis: {vision_data}"
-    else:
-        formatted_strings['vision'] = ""
-    return formatted_strings
-
-
-# --- FEATURE 1: Symptom Synthesizer Pipeline (v1.1) ---
-
-async def run_symptom_synthesis(user_query: str, image_input: Image.Image | None) -> str:
-    """The complete, asynchronous pipeline for the Symptom Synthesizer tab."""
-    if not user_query:
-        return "Please enter a symptom description or a medical question to begin."
 
-    # ==============================================================================
-    # STEP 1 (V1.1 UPGRADE): AI-Powered Query Correction (The "Medical Translator")
-    # ==============================================================================
-    correction_prompt = prompts.get_query_correction_prompt(user_query)
-    corrected_query = await gemini_handler.generate_text_response(correction_prompt)
-    if not corrected_query:
-        corrected_query = user_query # Fallback to original query if correction fails
+# (This function remains the same)
+def get_query_correction_prompt(user_text: str) -> str:
+    return f"""
+    You are an expert medical transcriptionist. Your task is to correct and clarify the following user query for a medical database search.
+    - Correct all spelling and grammatical errors.
+    - Translate colloquialisms or typos into proper medical terminology (e.g., "pin" -> "pain", "abdomian" -> "abdomen").
+    - Rephrase as a clear statement or question.
+    - Do not answer the question. Only return the corrected and clarified query.
+    User Query: "{user_text}"
+    Response:
+    """
 
-    # ==============================================================================
-    # STEP 2: AI-Powered Concept Extraction (now on the CLEANED query)
-    # ==============================================================================
-    term_prompt = prompts.get_term_extraction_prompt(corrected_query)
-    concepts_str = await gemini_handler.generate_text_response(term_prompt)
-    concepts = utils.safe_literal_eval(concepts_str)
-    if not isinstance(concepts, list) or not concepts:
-        concepts = [corrected_query]  # Fallback if parsing fails
+# (This function remains the same)
+def get_term_extraction_prompt(user_text: str) -> str:
+    return f"""
+    From the user's corrected query below, extract the most relevant medical concepts, symptoms, or conditions.
+    Return ONLY a Python-style list of strings.
+    User Text: "{user_text}"
+    Response:
+    """
 
-    # Use "OR" for a broader, more inclusive search across APIs
-    search_query = " OR ".join(f'"{c}"' for c in concepts)
 
+# ==============================================================================
+# V1.2 UPGRADE: The Symptom Synthesis prompt is now a "Narrative Briefing"
+# ==============================================================================
+def get_synthesis_prompt(user_query: str, concepts: list, pubmed_data: str, trials_data: str, fda_data: str, vision_analysis: str = "") -> str:
+    
     # ==============================================================================
-    # STEP 3: Massively Parallel Evidence Gathering
+    # CORRECTED LINE: The f-string logic is now handled correctly.
+    vision_section = f"## Analysis of Uploaded Image\n{vision_analysis}" if vision_analysis else ""
     # ==============================================================================
-    async with aiohttp.ClientSession() as session:
-        tasks = {
-            "pubmed": pubmed_client.search_pubmed(session, search_query, max_results=3),
-            "trials": clinicaltrials_client.find_trials(session, search_query, max_results=3),
-            "openfda": asyncio.gather(*(openfda_client.get_adverse_events(session, c, top_n=3) for c in concepts)),
-        }
-        if image_input:
-            tasks["vision"] = gemini_handler.analyze_image_with_text(
-                "In the context of the user query, analyze this image objectively. Describe visual features like color, shape, texture, and patterns. Do not diagnose or offer medical advice.", image_input
-            )
-        raw_results = await asyncio.gather(*tasks.values(), return_exceptions=True)
-        api_data = dict(zip(tasks.keys(), raw_results))
 
-    # ==============================================================================
-    # STEP 4: Data Formatting
-    # ==============================================================================
-    formatted_data = _format_api_data_for_prompt(api_data)
+    return f"""
+    You are Asclepius, an expert medical information analyst. Your task is to transform raw medical data into a coherent, insightful, and beautifully formatted narrative report for a user.
 
-    # ==============================================================================
-    # STEP 5: The Grand Synthesis
-    # ==============================================================================
-    synthesis_prompt = prompts.get_synthesis_prompt(
-        user_query=user_query, # Pass original query for context
-        concepts=concepts,
-        pubmed_data=formatted_data['pubmed'],
-        trials_data=formatted_data['trials'],
-        fda_data=formatted_data['openfda'],
-        vision_analysis=formatted_data['vision']
-    )
-    final_report = await gemini_handler.generate_text_response(synthesis_prompt)
+    **YOUR DIRECTIVES:**
+    1.  **START IMMEDIATELY with the provided mandatory disclaimer.** DO NOT add any other preamble, introduction, or disclaimer of your own. Your response must begin with "⚠️ IMPORTANT DISCLAIMER...".
+    2.  **WRITE A NARRATIVE, NOT A LIST.** Do not use "1.", "2.", "3." to structure the main report. Use Markdown headings (##) for each section.
+    3.  **SYNTHESIZE, DON'T JUST LIST.** For each section, provide a short introductory sentence that gives context, then present the data.
+    4.  **BE HELPFUL WHEN DATA IS EMPTY.** If a data source is empty, state that no specific data was found and then provide a brief, high-level overview of the concept from your general knowledge.
+
+    **REPORT STRUCTURE:**
+
+    ## Overview
+    (Start with a short, empathetic paragraph acknowledging the user's query about "{user_query}" and explaining that you have searched public health databases for information on the interpreted concepts: {concepts}.)
+
+    ## Insights from Medical Research
+    (Introduce this section by explaining you've looked for recent review articles on PubMed. Then, summarize the findings or state that none were found and give a general overview.)
+    {pubmed_data if pubmed_data else "No specific review articles were found on PubMed for this query."}
+
+    ## Current Clinical Trials
+    (Introduce this section by explaining these are active studies from ClinicalTrials.gov. Then, list the trials or state that none were found.)
+    {trials_data if trials_data else "No actively recruiting clinical trials were found matching this query."}
+
+    ## Related Drug & Safety Data
+    (Introduce this section by explaining this data comes from OpenFDA. Then, list the findings or state that none were found.)
+    {fda_data if fda_data else "No specific adverse event data was found for this query."}
+
+    {vision_section}
+
+    **Begin your report now. Adhere strictly to these directives.**
+    """
+
+
+# ==============================================================================
+# V1.2 UPGRADE: The Drug Interaction prompt is now an "Executive Safety Briefing"
+# ==============================================================================
+def get_drug_interaction_synthesis_prompt(drug_names: list[str], interaction_data: str, safety_data: str) -> str:
+    return f"""
+    You are a specialist AI focused on drug safety analysis. Your task is to act as a clear, cautious, and organized pharmacist, explaining raw API data to a user.
+
+    **YOUR DIRECTIVES:**
+    1.  **START IMMEDIATELY with the provided mandatory disclaimer.** DO NOT add any other preamble, introduction, or second disclaimer.
+    2.  **WRITE A HUMAN-READABLE BRIEFING.** Do not use sterile numbering ("1.", "2.", "3."). Use descriptive Markdown headings (##).
+    3.  **PROVIDE CONTEXT AND INSIGHT.** Your job is to explain what the data *means* in simple terms.
+
+    **BRIEFING STRUCTURE:**
+
+    ## Executive Summary
+    (Write a concise, 1-2 sentence summary of the most important findings. For example: "A review of {', '.join(drug_names)} found no direct drug-drug interactions, but did identify several commonly reported side effects for each medication." or "A potentially significant interaction was identified between Drug A and Drug B. Details are provided below.")
+
+    ## Drug-Drug Interaction Analysis
+    (If interactions exist, list them here. For each interaction, **explain the consequence in simple terms.** For example: "Taking these together may increase the risk of...". If none, state clearly: "No direct drug-drug interactions were found among the provided list of medications based on the data available.")
+    {interaction_data if interaction_data else "No direct drug-drug interactions were found."}
+
+    ## Individual Drug Safety Profiles
+    (Create a subsection for each drug using `### Drug Name`. Under each, summarize the data found in a user-friendly way.)
+    {safety_data if safety_data else "No individual safety profiles were found."}
     
-    # ==============================================================================
-    # STEP 6: Final Delivery
-    # ==============================================================================
-    return f"{prompts.DISCLAIMER}\n\n{final_report}"
-
-
-# --- FEATURE 2: Drug Interaction & Safety Analyzer Pipeline ---
-# (This function remains unchanged)
-async def run_drug_interaction_analysis(drug_list_str: str) -> str:
-    """The complete, asynchronous pipeline for the Drug Interaction Analyzer tab."""
-    if not drug_list_str:
-        return "Please enter a comma-separated list of medications."
-    drug_names = [name.strip() for name in drug_list_str.split(',') if name.strip()]
-    if len(drug_names) < 2:
-        return "Please enter at least two medications to check for interactions."
-    async with aiohttp.ClientSession() as session:
-        tasks = {
-            "interactions": rxnorm_client.run_interaction_check(drug_names),
-            "safety_profiles": asyncio.gather(*(openfda_client.get_safety_profile(session, name) for name in drug_names))
-        }
-        raw_results = await asyncio.gather(*tasks.values(), return_exceptions=True)
-        api_data = dict(zip(tasks.keys(), raw_results))
-    interaction_data = api_data.get('interactions', [])
-    if isinstance(interaction_data, Exception):
-        interaction_data = [{"error": str(interaction_data)}]
-    safety_profiles = api_data.get('safety_profiles', [])
-    if isinstance(safety_profiles, Exception):
-        safety_profiles = [{"error": str(safety_profiles)}]
-    safety_data_dict = dict(zip(drug_names, safety_profiles))
-    interaction_formatted = utils.format_list_as_markdown([str(i) for i in interaction_data]) if interaction_data else "No interactions found."
-    safety_formatted = "\n".join([f"Profile for {drug}: {profile}" for drug, profile in safety_data_dict.items()])
-    synthesis_prompt = prompts.get_drug_interaction_synthesis_prompt(
-        drug_names=drug_names,
-        interaction_data=interaction_formatted,
-        safety_data=safety_formatted
-    )
-    final_report = await gemini_handler.generate_text_response(synthesis_prompt)
-    return f"{prompts.DISCLAIMER}\n\n{final_report}"
+    **Begin your safety briefing now. Adhere strictly to these directives.**
+    """