Update app.py

app.py

@@ -34,16 +34,16 @@ WEAVIATE_URL="https://xbvlj5rpqyiswspww0tthq.c0.us-west3.gcp.weaviate.cloud"
 WEAVIATE_API_KEY="RU9acU1CYnNRTjY1S1ZFc18zNS9tQktaWlcwTzFEUjlscEVCUGF4YU5xRWx2MDhmTUtIdUhnOWdOTGVZPV92MjAw"
 DEEPINFRA_API_KEY="285LUJulGIprqT6hcPhiXtcrphU04FG4"
 DEEPINFRA_BASE_URL="https://api.deepinfra.com/v1/openai"
-
 # Initialize OpenAI client
+env = os.getenv("ENVIRONMENT", "production")
 openai = OpenAI(
     api_key=DEEPINFRA_API_KEY,
     base_url="https://api.deepinfra.com/v1/openai",
 )
-
 SESSION_ID = "default"
 
-# Chat Completion Helper
+pending_clarifications = {}
+
 def call_llm(model: str, messages: list[dict], temperature: float = 0.0, **kwargs) -> str:
     resp = openai.chat.completions.create(
         model=model,
@@ -53,66 +53,77 @@ def call_llm(model: str, messages: list[dict], temperature: float = 0.0, **kwarg
     )
     return resp.choices[0].message.content.strip()
 
-# Greeting Detection
 def is_greeting(text: str) -> bool:
     return bool(re.search(r"\b(hi|hello|hey|good (morning|afternoon|evening))\b", text, re.I))
 
-# Logging
-def _save_process_log(log_lines, filename=None):
-    logs_dir = os.path.join(os.path.dirname(__file__), "logs")
-    os.makedirs(logs_dir, exist_ok=True)
-    log_filename = filename or os.path.join(logs_dir, f"chat_session_{SESSION_ID}.txt")
-    try:
-        with open(log_filename, "a", encoding="utf-8") as f:
-            f.write("=" * 50 + "\n")
-            for line in log_lines:
-                f.write(str(line) + "\n\n")
-    except Exception as e:
-        print("Logging error:", e)
-
-# Main Process Function
-def process_query(query: str, first_turn: bool = False):
+def process_query(query: str, first_turn: bool = False, session_id: str = "default"):
     intro = ""
     process_log = []
 
+    if session_id in pending_clarifications:
+        if query.strip().lower() == "yes":
+            corrected_query = pending_clarifications.pop(session_id)
+            process_log.append(f"User confirmed: {corrected_query}")
+            return process_autism_pipeline(corrected_query, process_log, intro)
+        else:
+            pending_clarifications.pop(session_id)
+            redirect = "Hello I’m Wisal, an AI assistant developed by Compumacy AI, and a knowledgeable Autism specialist.\nIf you have any question related to autism please submit a question specifically about autism."
+            process_log.append("User rejected clarification.")
+            _save_process_log(process_log)
+            return redirect
+
     if first_turn and (not query or query.strip() == ""):
         intro = "Hello! I’m Wisal, an AI assistant developed by Compumacy AI, specializing in Autism Spectrum Disorders. How can I help you today?"
-        process_log.append("User: [empty or first turn]")
-        process_log.append(f"Wisal: {intro}")
+        process_log.append(intro)
         _save_process_log(process_log)
         return intro
 
-    process_log.append(f"User: {query}")
-
     if is_greeting(query):
         greeting = intro + "Hello! I’m Wisal, your AI assistant developed by Compumacy AI. How can I help you today?"
-        process_log.append(f"Wisal: {greeting}")
+        process_log.append(f"Greeting detected.\n{greeting}")
         _save_process_log(process_log)
         return greeting
 
     corrected_query = call_llm(
         model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_translation.format(query=query)}]
+        messages=[{"role": "user", "content": Prompt_template_translation.format(query=query)}],
+        reasoning_effort="none"
     )
     process_log.append(f"Corrected Query: {corrected_query}")
 
     relevance = call_llm(
         model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_relevance.format(corrected_query=corrected_query)}]
+        messages=[{"role": "user", "content": Prompt_template_relevance.format(corrected_query=corrected_query)}],
+        reasoning_effort="none"
     )
-    process_log.append(f"Relevance: {relevance}")
+    process_log.append(f"Relevance Check: {relevance}")
+
+    redirect_message = "Hello I’m Wisal, an AI assistant developed by Compumacy AI, and a knowledgeable Autism specialist.\nIf you have any question related to autism please submit a question specifically about autism."
+
+    if relevance.startswith("Hello I’m Wisal"):
+        clarification = f"Your query was not clearly related to autism. Do you mean:\n\"{corrected_query}\"\nIf yes, please confirm so I can help. If not:\n{redirect_message}"
+        pending_clarifications[session_id] = corrected_query
+        process_log.append(f"Clarification Prompted: {clarification}")
+        _save_process_log(process_log)
+        return clarification
+
     if relevance != "RELATED":
-        process_log.append(f"Wisal: {relevance}")
+        process_log.append("Query not autism-related.")
         _save_process_log(process_log)
-        return relevance
+        return intro + relevance
+
+    return process_autism_pipeline(corrected_query, process_log, intro)
 
+def process_autism_pipeline(corrected_query, process_log, intro):
     web_search_resp = asyncio.run(search_autism(corrected_query))
     web_answer = web_search_resp.get("answer", "")
-    process_log.append(f"Web Search Answer: {web_answer}")
+    process_log.append(f"Web Search: {web_answer}")
 
+    gen_prompt = Prompt_template_LLM_Generation.format(new_query=corrected_query)
     generated = call_llm(
         model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_LLM_Generation.format(new_query=corrected_query)}]
+        messages=[{"role": "user", "content": gen_prompt}],
+        reasoning_effort="none"
     )
     process_log.append(f"LLM Generated: {generated}")
 
@@ -120,66 +131,107 @@ def process_query(query: str, first_turn: bool = False):
     rag_contexts = rag_resp.get("answer", [])
     process_log.append(f"RAG Contexts: {rag_contexts}")
 
-    rag_text = "\n".join(f"[{i+1}] {c}" for i, c in enumerate(rag_contexts))
-    answers_list = f"[1] {generated}\n[2] {web_answer}\n{rag_text}"
+    answers_list = f"[1] {generated}\n[2] {web_answer}\n" + "\n".join(f"[{i+3}] {c}" for i, c in enumerate(rag_contexts))
+    rerank_prompt = Prompt_template_Reranker.format(new_query=corrected_query, answers_list=answers_list)
     reranked = call_llm(
         model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_Reranker.format(new_query=corrected_query, answers_list=answers_list)}]
+        messages=[{"role": "user", "content": rerank_prompt}],
+        reasoning_effort="none"
     )
     process_log.append(f"Reranked: {reranked}")
 
+    wisal_prompt = Prompt_template_Wisal.format(new_query=corrected_query, document=reranked)
     wisal = call_llm(
         model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_Wisal.format(new_query=corrected_query, document=reranked)}]
+        messages=[{"role": "user", "content": wisal_prompt}],
+        reasoning_effort="none"
     )
-    process_log.append(f"Wisal Final Answer: {wisal}")
+    process_log.append(f"Wisal Answer: {wisal}")
 
+    halluc_prompt = Prompt_template_Halluciations.format(
+        new_query=corrected_query,
+        answer=wisal,
+        document=generated
+    )
     halluc = call_llm(
         model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_Halluciations.format(new_query=corrected_query, answer=wisal, document=generated)}]
+        messages=[{"role": "user", "content": halluc_prompt}],
+        reasoning_effort="none"
     )
-    process_log.append(f"Hallucination Check: {halluc}")
-    score = int(halluc.split("Score: ")[1]) if "Score: " in halluc else 3
+    process_log.append(f"Hallucination Score: {halluc}")
+    score = int(halluc.split("Score: ")[-1]) if "Score: " in halluc else 3
 
     if score in (2, 3):
-        paraphrase = call_llm(
+        paraphrased = call_llm(
             model="Qwen/Qwen3-32B",
-            messages=[{"role": "user", "content": Prompt_template_paraphrasing.format(document=generated)}]
+            messages=[{"role": "user", "content": Prompt_template_paraphrasing.format(document=generated)}],
+            reasoning_effort="none"
         )
-        process_log.append(f"Paraphrased: {paraphrase}")
-        final_doc = call_llm(
+        wisal = call_llm(
             model="Qwen/Qwen3-32B",
-            messages=[{"role": "user", "content": Prompt_template_Wisal.format(new_query=corrected_query, document=paraphrase)}]
+            messages=[{"role": "user", "content": Prompt_template_Wisal.format(new_query=corrected_query, document=paraphrased)}],
+            reasoning_effort="none"
         )
-        process_log.append(f"Wisal with Paraphrase: {final_doc}")
-    else:
-        final_doc = wisal
+        process_log.append(f"Paraphrased Wisal: {wisal}")
 
     try:
-        detected_lang = langdetect.detect(query)
-    except Exception:
+        detected_lang = langdetect.detect(corrected_query)
+    except:
         detected_lang = "en"
+
     if detected_lang != "en":
         result = call_llm(
             model="Qwen/Qwen3-32B",
-            messages=[{"role": "user", "content": Prompt_template_Translate_to_original.format(query=query, document=final_doc)}]
+            messages=[{"role": "user", "content": Prompt_template_Translate_to_original.format(query=corrected_query, document=wisal)}],
+            reasoning_effort="none"
         )
         process_log.append(f"Translated Back: {result}")
     else:
-        result = final_doc
+        result = wisal
         process_log.append(f"Final Result: {result}")
 
-    process_log.append(f"Wisal: {result}")
     _save_process_log(process_log)
-    return result
+    return intro + result
+
+
+def _save_process_log(log_lines, filename="process_output.txt"):
+    import datetime
+    logs_dir = os.path.join(os.path.dirname(__file__), "logs")
+    os.makedirs(logs_dir, exist_ok=True)
+    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S_%f")
+    log_filename = os.path.join(logs_dir, f"log_{timestamp}.txt")
+    with open(log_filename, "w", encoding="utf-8") as f:
+        for line in log_lines:
+            f.write(str(line) + "\n\n")
+
+def _save_process_log(log_lines, filename="process_output.txt"):
+    import datetime
+    import os
+    # Ensure logs directory exists
+    logs_dir = os.path.join(os.path.dirname(__file__), "logs")
+    os.makedirs(logs_dir, exist_ok=True)
+    # Unique filename per question (timestamped)
+    timestamp = datetime.datetime.now().strftime("%Y%m%d_%H%M%S_%f")
+    log_filename = os.path.join(logs_dir, f"log_{timestamp}.txt")
+    try:
+        with open(log_filename, "w", encoding="utf-8") as f:
+            for line in log_lines:
+                f.write(str(line) + "\n\n")
+    except Exception as e:
+        pass
+
+
+# Gradio UI for main pipeline, RAG_Domain_know_doc, and User_Specific_Documents , Old_Document
+def main_pipeline_interface(query):
+    return process_query(query, first_turn=True)
 
-# Gradio Interface
 def main_pipeline_with_doc_and_history(query, doc_file, doc_type, history):
     response = main_pipeline_with_doc(query, doc_file, doc_type)
     updated_history = history + f"\nUser: {query}\nWisal: {response}\n"
     return response, updated_history
 
 def main_pipeline_with_doc(query, doc_file, doc_type):
+    # If no document, use main pipeline
     if doc_file is None or doc_type == "None":
         return process_query(query, first_turn=True)
 
@@ -189,15 +241,21 @@ def main_pipeline_with_doc(query, doc_file, doc_type):
 
     save_path = os.path.join(upload_dir, safe_filename)
 
+    # 💡 Check if doc_file is file-like (has `.read()`) or path-like (str or NamedString)
     if hasattr(doc_file, 'read'):
+        # File-like object
         file_bytes = doc_file.read()
     else:
+        # It's a path (NamedString), read from file path
         with open(str(doc_file), 'rb') as f:
             file_bytes = f.read()
 
+    # Save the file content
     with open(save_path, "wb") as f:
         f.write(file_bytes)
 
+
+    # Route to correct document handler
     if doc_type == "Knowledge Document":
         status = RAG_Domain_know_doc.ingest_file(save_path)
         answer = RAG_Domain_know_doc.answer_question(query)
@@ -212,6 +270,7 @@ def main_pipeline_with_doc(query, doc_file, doc_type):
         return f"[Old Document Uploaded]\n{status}\n\n{answer}"
     else:
         return "Invalid document type."
+    
 def pipeline_with_history(message, doc_file, doc_type, history):
     if not message.strip():
         return history, ""
@@ -219,7 +278,6 @@ def pipeline_with_history(message, doc_file, doc_type, history):
     history = history + [[message, response]]
     return history, ""
 
-
 with gr.Blocks(title="Wisal Chatbot", theme=gr.themes.Base()) as demo:
     gr.Markdown("# 🤖 Wisal: Autism AI Assistant")
 
@@ -245,17 +303,6 @@ with gr.Blocks(title="Wisal Chatbot", theme=gr.themes.Base()) as demo:
     clear_btn = gr.Button("Clear Chat")
     clear_btn.click(lambda: [], outputs=[chatbot])
 
-    # with gr.Tab("📘 Domain Knowledge RAG"):
-    #     RAG_Domain_know_doc.demo.render()
-
-    # with gr.Tab("📁 User-Specific Documents"):
-    #     User_Specific_Documents.demo.render()
-
-    # with gr.Tab("🕰️ Old Documents"):
-    #     Old_Document.demo.render()
 
 if __name__ == "__main__":
-    demo.launch(debug=True)
-
-    
-    
+    demo.launch(debug=True)