Update app.py

app.py

@@ -1,283 +1,288 @@
-import os
-import re
-import asyncio
-import gradio as gr
-import RAG_Domain_know_doc
-from web_search import search_autism
-from RAG import rag_autism
-from openai import OpenAI  # Corrected import
-from dotenv import load_dotenv
-import Old_Document 
-import User_Specific_Documents
-from prompt_template import (
-    Prompt_template_translation,
-    Prompt_template_LLM_Generation,
-    Prompt_template_Reranker,
-    Prompt_template_Wisal,
-    Prompt_template_Halluciations,
-    Prompt_template_paraphrasing,
-    Prompt_template_Translate_to_original,
-    Prompt_template_relevance,
-    Prompt_template_User_document_prompt
-)
-
-load_dotenv()
-DEEPINFRA_TOKEN = os.getenv("DEEPINFRA_API_KEY") or "285LUJulGIprqT6hcPhiXtcrphU04FG4"
-if not DEEPINFRA_TOKEN:
-    raise ValueError("DEEPINFRA_API_KEY is not set in .env file")
-
-# Initialize OpenAI client
-env = os.getenv("ENVIRONMENT", "production")
-openai = OpenAI(
-    api_key=DEEPINFRA_TOKEN,
-    base_url="https://api.deepinfra.com/v1/openai",
-)
-
-
-
-# Rest of your code remains unchanged
-# Helper to call chat completion synchronously
-def call_llm(model: str, messages: list[dict], temperature: float = 0.0, **kwargs) -> str:
-    resp = openai.chat.completions.create(
-        model=model,
-        messages=messages,
-        temperature=temperature,
-        **kwargs
-    )
-    return resp.choices[0].message.content.strip()
-
-# Basic 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))
-
-
-def process_query(query: str, first_turn: bool = False):
-    intro = ""
-    process_log = []
-
-    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(intro)
-        _save_process_log(process_log)
-        return intro
-
-    # ✅ Handle Yes/No replies
-    if query.strip().lower() == "no":
-        no_reply = (
-            "Hello, I’m Wisal, an AI assistant developed by Compumacy AI, "
-            "and a knowledgeable Autism specialist.\n"
-            "If you have any question related to autism, please submit a question specifically about autism."
-        )
-        process_log.append(f"User replied 'No'.\n{no_reply}")
-        _save_process_log(process_log)
-        return no_reply
-    elif query.strip().lower() == "yes":
-        process_log.append("User replied 'Yes'. Continuing system as normal.")
-
-
-    # 0: Handle simple greetings
-    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"Greeting detected.\n{greeting}")
-        _save_process_log(process_log)
-        return greeting
-
-    # 1: Translation & Rephrasing
-    corrected_query = call_llm(
-        model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_translation.format(query=query)}],
-        reasoning_effort="none"
-    )
-    process_log.append(f"Corrected Query: {corrected_query}")
-
-    # 2: Relevance Check
-    relevance = call_llm(
-        model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": Prompt_template_relevance.format(corrected_query=corrected_query)}],
-        reasoning_effort="none"
-    )
-    process_log.append(f"Relevance: {relevance}")
-    if relevance != "RELATED":
-        process_log.append(f"Query not related. Returning: {relevance}")
-        _save_process_log(process_log)
-        return intro + relevance
-
-    # Step 3: Web Search
-    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}")
-
-    # Step 4: LLM Generation
-    gen_prompt = Prompt_template_LLM_Generation.format(new_query=corrected_query)
-    generated = call_llm(
-        model="Qwen/Qwen3-32B",
-        messages=[{"role": "user", "content": gen_prompt}],
-        reasoning_effort="none"
-    )
-    process_log.append(f"LLM Generated: {generated}")
-
-    # Step 5: RAG
-    rag_resp = asyncio.run(rag_autism(corrected_query, top_k=3))
-    rag_contexts = rag_resp.get("answer", [])
-    process_log.append(f"RAG Contexts: {rag_contexts}")
-
-    # 6) Reranking (now across 3 candidates)
-    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}"
-    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":rerank_prompt}],
-        reasoning_effort="none"
-    )
-    process_log.append(f"Reranked: {reranked}")
-
-    # 7) Wisal final‐answer generation
-    wisal_prompt = Prompt_template_Wisal.format(
-        new_query=corrected_query,
-        document=reranked  # use reranked output here
-    )
-    wisal = call_llm(
-        model="Qwen/Qwen3-32B",
-        messages=[{"role":"user","content":wisal_prompt}],
-        reasoning_effort="none"
-    )
-    process_log.append(f"Wisal Final Answer: {wisal}")
-
-    # 8) Hallucination Check
-    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": halluc_prompt}],
-        reasoning_effort="none"
-    )
-    process_log.append(f"Hallucination Check: {halluc}")
-    score = int(halluc.split("Score: ")[1]) if "Score: " in halluc else 3
-
-    # 9) Paraphrase if needed
-    if score in (2, 3):
-        paraphrase = call_llm(
-            model="Qwen/Qwen3-32B",
-            messages=[{"role": "user", "content": Prompt_template_paraphrasing.format(document=generated)}],
-            reasoning_effort="none"
-        )
-        process_log.append(f"Paraphrased: {paraphrase}")
-        context_prompt = Prompt_template_Wisal.format(new_query=corrected_query, document=paraphrase)
-        final_doc = call_llm(
-            model="Qwen/Qwen3-32B",
-            messages=[{"role": "user", "content": context_prompt}],
-            reasoning_effort="none"
-        )
-        process_log.append(f"Wisal with Paraphrase: {final_doc}")
-    else:
-        final_doc = wisal
-
-    # 10) Translate back if needed (improved: only if input is not English)
-    import langdetect
-    try:
-        detected_lang = langdetect.detect(query)
-    except Exception:
-        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)}],
-            reasoning_effort="none"
-        )
-        process_log.append(f"Translated Back: {result}")
-    else:
-        result = final_doc
-        process_log.append(f"Final Result: {result}")
-
-    _save_process_log(process_log)
-    return intro + result
-# Utility to save process log to a txt file
-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)
-
-
-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)
-
-    safe_filename = os.path.basename(getattr(doc_file, 'name', str(doc_file)))
-    upload_dir = os.path.join(os.path.dirname(__file__), "uploaded_docs")
-    os.makedirs(upload_dir, exist_ok=True)
-
-    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)
-        return f"[Knowledge Document Uploaded]\n{status}\n\n{answer}"
-    elif doc_type == "User-Specific Document":
-        status = User_Specific_Documents.ingest_file(save_path)
-        answer = User_Specific_Documents.answer_question(query)
-        return f"[User-Specific Document Uploaded]\n{status}\n\n{answer}"
-    elif doc_type == "Old Document":
-        status = Old_Document.ingest_file(save_path)
-        answer = Old_Document.answer_question(query)
-        return f"[Old Document Uploaded]\n{status}\n\n{answer}"
-    else:
-        return "Invalid document type."
-
-with gr.Blocks(title="Wisal Main Pipeline & RAG") as demo:
-    gr.Markdown("## Wisal: Autism AI Assistant (Main Pipeline)")
-    with gr.Tab("Main Pipeline"):
-        q = gr.Textbox(placeholder="Your question...", lines=2, label="Ask Wisal")
-        doc_file = gr.File(label="Optional: Upload Document (PDF, DOCX, TXT)")
-        doc_type = gr.Radio(["None", "Knowledge Document", "User-Specific Document", "Old Document"], value="None", label="Document Type")
-        btn = gr.Button("Submit")
-        out = gr.Textbox(label="Wisal Answer", lines=8, interactive=False)
-        btn.click(fn=main_pipeline_with_doc, inputs=[q, doc_file, doc_type], outputs=out)
-    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)
-
+import os
+import re
+import asyncio
+import gradio as gr
+import RAG_Domain_know_doc
+from web_search import search_autism
+from RAG import rag_autism
+from openai import OpenAI  # Corrected import
+from dotenv import load_dotenv
+import Old_Document 
+import User_Specific_Documents
+from prompt_template import (
+    Prompt_template_translation,
+    Prompt_template_LLM_Generation,
+    Prompt_template_Reranker,
+    Prompt_template_Wisal,
+    Prompt_template_Halluciations,
+    Prompt_template_paraphrasing,
+    Prompt_template_Translate_to_original,
+    Prompt_template_relevance,
+    Prompt_template_User_document_prompt
+)
+
+GEMINI_API_KEY="AIzaSyCUCivstFpC9pq_jMHMYdlPrmh9Bx97dFo"
+TAVILY_API_KEY="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"
+OPENAI_API_KEY="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"
+QDRANT_API_KEY="eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJhY2Nlc3MiOiJtIiwiZXhwIjoxNzUxMDUxNzg4fQ.I9J-K7OM0BtcNKgj2d4uVM8QYAHYfFCVAyP4rlZkK2E"
+QDRANT_URL="https://6a3aade6-e8ad-4a6c-a579-21f5af90b7e8.us-east4-0.gcp.cloud.qdrant.io"
+OPENAI_API_KEY="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"
+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_TOKEN,
+    base_url="https://api.deepinfra.com/v1/openai",
+)
+
+
+
+# Rest of your code remains unchanged
+# Helper to call chat completion synchronously
+def call_llm(model: str, messages: list[dict], temperature: float = 0.0, **kwargs) -> str:
+    resp = openai.chat.completions.create(
+        model=model,
+        messages=messages,
+        temperature=temperature,
+        **kwargs
+    )
+    return resp.choices[0].message.content.strip()
+
+# Basic 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))
+
+
+def process_query(query: str, first_turn: bool = False):
+    intro = ""
+    process_log = []
+
+    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(intro)
+        _save_process_log(process_log)
+        return intro
+
+    # ✅ Handle Yes/No replies
+    if query.strip().lower() == "no":
+        no_reply = (
+            "Hello, I’m Wisal, an AI assistant developed by Compumacy AI, "
+            "and a knowledgeable Autism specialist.\n"
+            "If you have any question related to autism, please submit a question specifically about autism."
+        )
+        process_log.append(f"User replied 'No'.\n{no_reply}")
+        _save_process_log(process_log)
+        return no_reply
+    elif query.strip().lower() == "yes":
+        process_log.append("User replied 'Yes'. Continuing system as normal.")
+
+
+    # 0: Handle simple greetings
+    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"Greeting detected.\n{greeting}")
+        _save_process_log(process_log)
+        return greeting
+
+    # 1: Translation & Rephrasing
+    corrected_query = call_llm(
+        model="Qwen/Qwen3-32B",
+        messages=[{"role": "user", "content": Prompt_template_translation.format(query=query)}],
+        reasoning_effort="none"
+    )
+    process_log.append(f"Corrected Query: {corrected_query}")
+
+    # 2: Relevance Check
+    relevance = call_llm(
+        model="Qwen/Qwen3-32B",
+        messages=[{"role": "user", "content": Prompt_template_relevance.format(corrected_query=corrected_query)}],
+        reasoning_effort="none"
+    )
+    process_log.append(f"Relevance: {relevance}")
+    if relevance != "RELATED":
+        process_log.append(f"Query not related. Returning: {relevance}")
+        _save_process_log(process_log)
+        return intro + relevance
+
+    # Step 3: Web Search
+    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}")
+
+    # Step 4: LLM Generation
+    gen_prompt = Prompt_template_LLM_Generation.format(new_query=corrected_query)
+    generated = call_llm(
+        model="Qwen/Qwen3-32B",
+        messages=[{"role": "user", "content": gen_prompt}],
+        reasoning_effort="none"
+    )
+    process_log.append(f"LLM Generated: {generated}")
+
+    # Step 5: RAG
+    rag_resp = asyncio.run(rag_autism(corrected_query, top_k=3))
+    rag_contexts = rag_resp.get("answer", [])
+    process_log.append(f"RAG Contexts: {rag_contexts}")
+
+    # 6) Reranking (now across 3 candidates)
+    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}"
+    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":rerank_prompt}],
+        reasoning_effort="none"
+    )
+    process_log.append(f"Reranked: {reranked}")
+
+    # 7) Wisal final‐answer generation
+    wisal_prompt = Prompt_template_Wisal.format(
+        new_query=corrected_query,
+        document=reranked  # use reranked output here
+    )
+    wisal = call_llm(
+        model="Qwen/Qwen3-32B",
+        messages=[{"role":"user","content":wisal_prompt}],
+        reasoning_effort="none"
+    )
+    process_log.append(f"Wisal Final Answer: {wisal}")
+
+    # 8) Hallucination Check
+    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": halluc_prompt}],
+        reasoning_effort="none"
+    )
+    process_log.append(f"Hallucination Check: {halluc}")
+    score = int(halluc.split("Score: ")[1]) if "Score: " in halluc else 3
+
+    # 9) Paraphrase if needed
+    if score in (2, 3):
+        paraphrase = call_llm(
+            model="Qwen/Qwen3-32B",
+            messages=[{"role": "user", "content": Prompt_template_paraphrasing.format(document=generated)}],
+            reasoning_effort="none"
+        )
+        process_log.append(f"Paraphrased: {paraphrase}")
+        context_prompt = Prompt_template_Wisal.format(new_query=corrected_query, document=paraphrase)
+        final_doc = call_llm(
+            model="Qwen/Qwen3-32B",
+            messages=[{"role": "user", "content": context_prompt}],
+            reasoning_effort="none"
+        )
+        process_log.append(f"Wisal with Paraphrase: {final_doc}")
+    else:
+        final_doc = wisal
+
+    # 10) Translate back if needed (improved: only if input is not English)
+    import langdetect
+    try:
+        detected_lang = langdetect.detect(query)
+    except Exception:
+        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)}],
+            reasoning_effort="none"
+        )
+        process_log.append(f"Translated Back: {result}")
+    else:
+        result = final_doc
+        process_log.append(f"Final Result: {result}")
+
+    _save_process_log(process_log)
+    return intro + result
+# Utility to save process log to a txt file
+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)
+
+
+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)
+
+    safe_filename = os.path.basename(getattr(doc_file, 'name', str(doc_file)))
+    upload_dir = os.path.join(os.path.dirname(__file__), "uploaded_docs")
+    os.makedirs(upload_dir, exist_ok=True)
+
+    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)
+        return f"[Knowledge Document Uploaded]\n{status}\n\n{answer}"
+    elif doc_type == "User-Specific Document":
+        status = User_Specific_Documents.ingest_file(save_path)
+        answer = User_Specific_Documents.answer_question(query)
+        return f"[User-Specific Document Uploaded]\n{status}\n\n{answer}"
+    elif doc_type == "Old Document":
+        status = Old_Document.ingest_file(save_path)
+        answer = Old_Document.answer_question(query)
+        return f"[Old Document Uploaded]\n{status}\n\n{answer}"
+    else:
+        return "Invalid document type."
+
+with gr.Blocks(title="Wisal Main Pipeline & RAG") as demo:
+    gr.Markdown("## Wisal: Autism AI Assistant (Main Pipeline)")
+    with gr.Tab("Main Pipeline"):
+        q = gr.Textbox(placeholder="Your question...", lines=2, label="Ask Wisal")
+        doc_file = gr.File(label="Optional: Upload Document (PDF, DOCX, TXT)")
+        doc_type = gr.Radio(["None", "Knowledge Document", "User-Specific Document", "Old Document"], value="None", label="Document Type")
+        btn = gr.Button("Submit")
+        out = gr.Textbox(label="Wisal Answer", lines=8, interactive=False)
+        btn.click(fn=main_pipeline_with_doc, inputs=[q, doc_file, doc_type], outputs=out)
+    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)
+