Update app.py

app.py

@@ -1,6 +1,6 @@
 """
 Jajabor – SEBA Assamese Class 10 Tutor (Free-tier CPU-ready)
-Fixed version with correct Gradio version and improved error handling
+Fixed version with Gradio compatibility fixes
 """
 
 import os
@@ -32,9 +32,9 @@ USE_HF_INFERENCE = False
 LLM_LOCAL_NAME = "google/flan-t5-small"
 LLM_MAX_TOKENS = 128
 
-CHUNK_SIZE = 600
-CHUNK_OVERLAP = 120
-TOP_K = 5
+CHUNK_SIZE = 400  # Reduced for better performance
+CHUNK_OVERLAP = 80
+TOP_K = 3  # Reduced for faster retrieval
 
 # -------------------- DATABASE --------------------
 def init_db(path=DB_PATH):
@@ -139,8 +139,9 @@ def load_all_pdfs(pdf_dir: str):
             path = os.path.join(pdf_dir, fname)
             print("Reading:", path)
             text = extract_text_from_pdf(path)
-            texts.append(text)
-            metas.append({"source": fname})
+            if text.strip():
+                texts.append(text)
+                metas.append({"source": fname})
     return texts, metas
 
 def split_text(text: str, chunk_size=CHUNK_SIZE, overlap=CHUNK_OVERLAP):
@@ -164,7 +165,7 @@ embedding_model = SentenceTransformer(EMBEDDING_MODEL_NAME)
 
 print("Loading PDFs from", PDF_DIR)
 all_texts, all_metas = load_all_pdfs(PDF_DIR)
-print("Number of PDFs:", len(all_texts))
+print("Number of PDFs with content:", len(all_texts))
 
 corpus_chunks = []
 corpus_metas = []
@@ -178,7 +179,7 @@ index = None
 if len(corpus_chunks) > 0:
     print("Encoding chunks...")
     try:
-        embs = embedding_model.encode(corpus_chunks, batch_size=32, show_progress_bar=False).astype("float32")
+        embs = embedding_model.encode(corpus_chunks, batch_size=16, show_progress_bar=False).astype("float32")
         dim = embs.shape[1]
         index = faiss.IndexFlatL2(dim)
         index.add(embs)
@@ -217,49 +218,54 @@ llm_pipe = None
 try:
     tokenizer = AutoTokenizer.from_pretrained(LLM_LOCAL_NAME)
     model = AutoModelForSeq2SeqLM.from_pretrained(LLM_LOCAL_NAME)
-    llm_pipe = pipeline("text2text-generation", model=model, tokenizer=tokenizer, device=-1)  # CPU
-    print("Local LLM loaded.")
+    llm_pipe = pipeline(
+        "text2text-generation", 
+        model=model, 
+        tokenizer=tokenizer, 
+        device=-1,  # CPU
+        torch_dtype="auto"
+    )
+    print("✅ Local LLM loaded successfully")
 except Exception as e:
     print("Failed to load local LLM:", e)
     llm_pipe = None
 
-SYSTEM_PROMPT = """
-You are "Jajabor", an expert SEBA Assamese tutor for Class 10.
-Always prefer to answer in Assamese. If the student clearly asks for English, you may reply in English.
-
-Rules:
-- Use ONLY the given textbook context.
-- If you are not sure, say: "এই প্ৰশ্নটো পাঠ্যপুথিৰ অংশত স্পষ্টকৈ নাই, সেয়েহে মই নিশ্চিত নহয়।"
-- বোঝাপৰা সহজ ভাষাত ব্যাখ্যা কৰা, উদাহৰণ দিয়ক।
-- If it is a maths question, explain step-by-step clearly.
-"""
+SYSTEM_PROMPT = """You are "Jajabor", an expert SEBA Assamese tutor for Class 10.
+Answer in Assamese unless the student asks for English.
+Use the textbook context provided. If unsure, say you don't know.
+Explain simply with examples."""
 
 def build_rag_prompt(context_blocks, question, chat_history):
     ctx = ""
     for i, block in enumerate(context_blocks, start=1):
         src = block["meta"].get("source", "textbook")
-        ctx += f"\n[Context {i} – {src}]\n{block['text']}\n"
+        ctx += f"[Context {i} - {src}]\n{block['text']}\n\n"
 
     hist = ""
-    for role, msg in chat_history[-4:]:  # Keep last 4 exchanges
-        hist += f"{role}: {msg}\n"
+    for u, a in chat_history[-3:]:  # Last 3 exchanges
+        if u:
+            hist += f"Student: {u}\n"
+        if a:
+            hist += f"Tutor: {a}\n"
 
     prompt = f"""{SYSTEM_PROMPT}
 
-পূৰ্বৰ বাৰ্তাসমূহ:
+Previous conversation:
 {hist}
 
-সদস্যৰ প্ৰশ্ন:
+Student's question:
 {question}
 
-সম্পৰ্কিত পাঠ্যপুথিৰ অংশ:
+Textbook content:
 {ctx}
 
-এতিয়া একেদম সহায়ক আৰু বুজিবলৈ সহজ উত্তৰ দিয়া।
-"""
+Provide a helpful, easy-to-understand answer in Assamese:"""
     return prompt
 
 def llm_answer_with_rag(question: str, chat_history):
+    if not question.strip():
+        return "অনুগ্ৰহ কৰি এটা প্ৰশ্ন সোধক।"
+    
     retrieved = rag_search(question, TOP_K)
     if not retrieved:
         return "মই এই প্ৰশ্নৰ উত্তৰ দিবলৈ প্ৰয়োজনীয় তথ্য বিচাৰি পোৱা নাই। দয়া কৰি নিশ্চিত কৰক যে আপোনাৰ পাঠ্যপুথিৰ PDF ফাইলসমূহ সঠিকভাৱে আপলোড কৰা হৈছে।"
@@ -270,15 +276,23 @@ def llm_answer_with_rag(question: str, chat_history):
         return "AI মডেল ল'ড হোৱা নাই। দয়া কৰি পুনৰ চেষ্টা কৰক।"
     
     try:
-        out = llm_pipe(prompt, max_new_tokens=LLM_MAX_TOKENS, do_sample=False)
+        out = llm_pipe(
+            prompt, 
+            max_new_tokens=LLM_MAX_TOKENS, 
+            do_sample=False,
+            temperature=0.3
+        )
         if isinstance(out, list) and len(out) > 0:
-            if "generated_text" in out[0]:
+            if hasattr(out[0], 'get') and "generated_text" in out[0]:
                 return out[0]["generated_text"]
-            return str(out[0])
-        return str(out)
+            elif isinstance(out[0], str):
+                return out[0]
+            else:
+                return str(out[0])
+        return "উত্তৰ তৈয়াৰ কৰোঁতে সমস্যা হ'ল।"
     except Exception as e:
         print("LLM generation error:", e)
-        return f"উত্তৰ তৈয়াৰ কৰোঁতে সমস্যা: {e}"
+        return f"উত্তৰ তৈয়াৰ কৰোঁতে ত্ৰুটি: {str(e)}"
 
 # -------------------- OCR + Math helpers --------------------
 def ocr_from_image(img_path: str):
@@ -287,7 +301,7 @@ def ocr_from_image(img_path: str):
     try:
         img = Image.open(img_path)
         img = img.convert("RGB")
-        text = pytesseract.image_to_string(img, lang="eng+asm")
+        text = pytesseract.image_to_string(img, lang="eng")
         return text.strip()
     except Exception as e:
         print("OCR error:", e)
@@ -297,41 +311,52 @@ def is_likely_math(text: str) -> bool:
     if not text:
         return False
     math_chars = set("0123456789+-*/=^()%")
-    if any(ch in text for ch in math_chars):
+    text_chars = set(text)
+    if math_chars.intersection(text_chars):
         return True
-    math_kws = ["গণিত", "সমীকৰণ", "উদাহৰণ", "প্ৰশ্ন", "বীজগণিত", "solve", "equation", "math"]
+    math_kws = ["গণিত", "সমীকৰণ", "উদাহৰণ", "প্ৰশ্ন", "বীজগণিত", "solve", "equation", "math", "calculate"]
     return any(k in text.lower() for k in math_kws)
 
 def solve_math_expression(expr: str):
     try:
-        expr = expr.replace("^", "**")
-        if "=" in expr:
-            left, right = expr.split("=", 1)
-            left_s = sp.sympify(left.strip())
-            right_s = sp.sympify(right.strip())
-            eq = sp.Eq(left_s, right_s)
-            sol = sp.solve(eq)
-            explanation = f"সমীকৰণ: {eq}\n\nসমাধান: {sol}"
+        # Clean the expression
+        expr = expr.strip()
+        expr = expr.replace('^', '**')
+        
+        if '=' in expr:
+            parts = expr.split('=')
+            if len(parts) == 2:
+                left = sp.sympify(parts[0].strip())
+                right = sp.sympify(parts[1].strip())
+                equation = sp.Eq(left, right)
+                solutions = sp.solve(equation)
+                
+                if solutions:
+                    solution_str = f"সমীকৰণ: {equation}\n\nসমাধান: x = {solutions[0]}"
+                    if len(solutions) > 1:
+                        solution_str += f"\nবা x = {solutions[1]}"
+                    return solution_str
+                else:
+                    return "কোনো সমাধান পোৱা নগ'ল।"
         else:
-            expr_s = sp.sympify(expr)
-            simp = sp.simplify(expr_s)
-            explanation = f"প্ৰকাশ: {expr}\n\nসৰলীকৃত: {simp}"
-        return explanation
+            # Just simplify the expression
+            expr_sym = sp.sympify(expr)
+            simplified = sp.simplify(expr_sym)
+            return f"প্ৰকাশ: {expr}\n\nসৰলীকৃত: {simplified}"
+            
     except Exception as e:
-        return f"গণিত সমাধানত সমস্যা: {e}"
-
-def speech_to_text(audio):
-    return ""  # Stub for future implementation
-
-def text_to_speech(text: str):
-    return None  # Stub for future implementation
+        return f"গণিত সমাধানত সমস্যা: {str(e)}\nদয়া কৰি স্পষ্টকৈ লিখক, যেনে: 2*x + 3 = 7"
 
 # -------------------- Chat logic --------------------
-def login_user(username, user_state):
+def login_user(username):
     username = (username or "").strip()
     if not username:
-        return user_state, "⚠️ অনুগ্ৰহ কৰি প্ৰথমে লগিনৰ বাবে এটা নাম লিখক।"
+        return {}, "⚠️ অনুগ্ৰহ কৰি প্ৰথমে লগিনৰ বাবে এটা নাম লিখক।"
+    
     user_id = get_or_create_user(username)
+    if not user_id:
+        return {}, "⚠️ লগিন কৰোঁতে সমস্যা হ'ল।"
+    
     user_state = {"username": username, "user_id": user_id}
     total, math_count = get_user_stats(user_id)
     stats = (
@@ -345,142 +370,169 @@ def chat_logic(text_input, image_input, chat_history, user_state):
     if chat_history is None:
         chat_history = []
 
+    # Check if user is logged in
     if not user_state or not user_state.get("user_id"):
-        sys_msg = "⚠️ প্ৰথমে ওপৰত আপোনাৰ নাম লিখি **Login / লগিন** টিপক।"
-        chat_history.append([text_input or "", sys_msg])
-        return chat_history, user_state, None
+        chat_history.append([text_input or "", "⚠️ প্ৰথমে ওপৰত আপোনাৰ নাম লিখি **Login / লগিন** টিপক।"])
+        return chat_history, user_state
 
     user_id = user_state["user_id"]
     final_query_parts = []
 
     # Process image OCR
-    ocr_text = ""
     if image_input is not None:
         ocr_text = ocr_from_image(image_input)
         if ocr_text:
-            final_query_parts.append(f"ছবিৰ পৰা পাঠ: {ocr_text}")
+            final_query_parts.append(f"[ছবিৰ পাঠ] {ocr_text}")
 
-    if text_input:
-        final_query_parts.append(text_input)
+    if text_input and text_input.strip():
+        final_query_parts.append(text_input.strip())
 
     if not final_query_parts:
-        sys_msg = "⚠️ অনুগ্ৰহ কৰি প্ৰশ্ন লিখক, কিম্বা ছবি আপলোড কৰক।"
-        chat_history.append(["", sys_msg])
-        return chat_history, user_state, None
+        chat_history.append(["", "⚠️ অনুগ্ৰহ কৰি প্ৰশ্ন লিখক, কিম্বা ছবি আপলোড কৰক।"])
+        return chat_history, user_state
 
     full_query = "\n".join(final_query_parts)
 
-    # Convert chat history to conversation format
-    conv = []
-    for u, b in chat_history:
-        if u and u.strip():
-            conv.append(("Student", u.strip()))
-        if b and b.strip():
-            conv.append(("Tutor", b.strip()))
-
     is_math = is_likely_math(full_query)
 
     if is_math:
         math_answer = solve_math_expression(full_query)
-        combined_question = (
-            full_query + "\n\nগণিত সমাধান:\n" + math_answer + 
-            "\n\nঅনুগ্ৰহ কৰি শ্রেণী ১০ ৰ শিক্ষাৰ্থীৰ বাবে সহজ ভাষাত ব্যাখ্যা কৰক।"
-        )
-        final_answer = llm_answer_with_rag(combined_question, conv)
+        # Combine math solution with request for explanation
+        combined_question = f"{full_query}\n\nগণিত সমাধান:\n{math_answer}\n\nঅনুগ্ৰহ কৰি ইয়াক সহজ ভাষাত ব্যাখ্যা কৰক:"
+        final_answer = llm_answer_with_rag(combined_question, chat_history)
     else:
-        final_answer = llm_answer_with_rag(full_query, conv)
+        final_answer = llm_answer_with_rag(full_query, chat_history)
 
     log_interaction(user_id, full_query, final_answer, is_math)
     
-    display_question = text_input or ocr_text or "(ছবিৰ প্ৰশ্ন)"
+    display_question = text_input or "[ছবিৰ প্ৰশ্ন]"
     chat_history.append([display_question, final_answer])
 
-    return chat_history, user_state, None
+    return chat_history, user_state
 
-# -------------------- Gradio UI --------------------
-with gr.Blocks(title=APP_NAME, css="""
-.stats-box { background: #f0f8ff; padding: 10px; border-radius: 5px; }
-""") as demo:
-    gr.Markdown(
-        f"""
-        # 🧭 {APP_NAME}
-
-        - SEBA Class 10 PDFs upload to `pdfs/class10` folder
-        - Text + Image (OCR) input support  
-        - Math step-by-step solutions
-        - User login + progress tracking
-        """
-    )
-
-    user_state = gr.State({})
+def clear_chat():
+    return [], None
 
+# -------------------- Gradio UI --------------------
+with gr.Blocks(
+    title=APP_NAME,
+    css="""
+    .stats-box { 
+        background: #f0f8ff; 
+        padding: 15px; 
+        border-radius: 8px; 
+        border: 1px solid #d1e7ff;
+        margin-bottom: 15px;
+    }
+    .login-section {
+        background: #f8f9fa;
+        padding: 15px;
+        border-radius: 8px;
+        margin-bottom: 15px;
+    }
+    """
+) as demo:
+    gr.Markdown(f"# 🧭 {APP_NAME}")
+    
+    gr.Markdown("""
+    - SEBA Class 10 PDFs upload to `pdfs/class10` folder
+    - Text + Image (OCR) input support  
+    - Math step-by-step solutions
+    - User login + progress tracking
+    """)
+
+    # Use a simpler state management approach
+    user_state = gr.State(value={})
+    
     with gr.Row():
         with gr.Column(scale=1):
-            gr.Markdown("### 👤 লগিন")
-            username_inp = gr.Textbox(
-                label="নাম / ইউজাৰ আইডি",
-                placeholder="উদাহৰণ: abu10, student01 ..."
-            )
-            login_btn = gr.Button("✅ Login / লগিন")
-            stats_md = gr.Markdown("এতিয়ালৈকে লগিন হোৱা নাই।", elem_classes="stats-box")
-
-            gr.Markdown(
-                """
-                ### 💡 টিপছ
-                - "ক্লাছ ১০ গণিত: উদাহৰণ ৩.১ প্ৰশ্ন ২" – এই ধৰণৰ প্ৰশ্ন ভাল
-                - ফটো আপলোড কৰিলে টেক্স্টটো OCR কৰি পঢ়িব চেষ্টা কৰা হয়
-                - সম্ভৱ হলে প্ৰশ্নটো অসমীয়াত সোধক 🙂
-                """
-            )
+            with gr.Group(elem_classes="login-section"):
+                gr.Markdown("### 👤 লগিন")
+                username_inp = gr.Textbox(
+                    label="নাম / ইউজাৰ আইডি",
+                    placeholder="উদাহৰণ: abu10, student01 ...",
+                    max_lines=1
+                )
+                login_btn = gr.Button("✅ Login / লগিন", variant="primary")
+                stats_md = gr.Markdown("এতিয়ালৈকে লগিন হোৱা নাই।", elem_classes="stats-box")
+
+            gr.Markdown("""
+            ### 💡 টিপছ
+            - "ক্লাছ ১০ গণিত: উদাহৰণ ৩.১ প্ৰশ্ন ২" – এই ধৰণৰ প্ৰশ্ন ভাল
+            - ফটো আপলোড কৰিলে টেক্স্টটো OCR কৰি পঢ়িব চেষ্টা কৰা হয়
+            - সম্ভৱ হলে প্ৰশ্নটো অসমীয়াত সোধক 🙂
+            """)
 
         with gr.Column(scale=3):
-            chat = gr.Chatbot(label="জাজাবৰ সৈতে কথোপকথন", height=500)
-            text_inp = gr.Textbox(
-                label="আপোনাৰ প্ৰশ্ন লিখক",
-                placeholder='উদাহৰণ: "ক্লাছ ১০ অসমীয়া: অনুচ্ছেদ পাঠ ১ ৰ মূল বিষয় কি?"',
-                lines=2,
+            chatbot = gr.Chatbot(
+                label="জাজাবৰ সৈতে কথোপকথন",
+                height=500,
+                show_copy_button=True
             )
-
+            
+            with gr.Row():
+                text_inp = gr.Textbox(
+                    label="আপোনাৰ প্ৰশ্ন লিখক",
+                    placeholder='উদাহৰণ: "ক্লাছ ১০ অসমীয়া: অনুচ্ছেদ পাঠ ১ ৰ মূল বিষয় কি?"',
+                    lines=2,
+                    scale=4
+                )
+            
             with gr.Row():
-                image_inp = gr.Image(label="📷 প্ৰশ্নৰ ছবি (Optional)", type="filepath")
+                image_inp = gr.Image(
+                    label="📷 প্ৰশ্নৰ ছবি (Optional)",
+                    type="filepath",
+                    scale=3
+                )
                 
             with gr.Row():
-                ask_btn = gr.Button("🤖 জাজাবৰক সোধক")
-                clear_btn = gr.Button("🧹 পৰিষ্কাৰ কৰক")
+                ask_btn = gr.Button("🤖 জাজাবৰক সোধক", variant="primary", scale=2)
+                clear_btn = gr.Button("🧹 পৰিষ্কাৰ কৰক", variant="secondary", scale=1)
 
-    # Login handler
+    # Event handlers
     login_btn.click(
         login_user,
-        inputs=[username_inp, user_state],
-        outputs=[user_state, stats_md],
+        inputs=[username_inp],
+        outputs=[user_state, stats_md]
     )
-
-    # Chat handler
+    
+    # Chat function - simplified
+    def process_chat(text, image, history, state):
+        return chat_logic(text, image, history, state)
+    
     ask_btn.click(
-        chat_logic,
-        inputs=[text_inp, image_inp, chat, user_state],
-        outputs=[chat, user_state, image_inp],
-    ).then(
-        lambda: "", None, text_inp
+        process_chat,
+        inputs=[text_inp, image_inp, chatbot, user_state],
+        outputs=[chatbot, user_state]
     ).then(
-        lambda: None, None, image_inp
+        lambda: ("", None),
+        outputs=[text_inp, image_inp]
     )
-
-    # Text submit handler
+    
     text_inp.submit(
-        chat_logic,
-        inputs=[text_inp, image_inp, chat, user_state],
-        outputs=[chat, user_state, image_inp],
-    ).then(
-        lambda: "", None, text_inp
+        process_chat,
+        inputs=[text_inp, image_inp, chatbot, user_state],
+        outputs=[chatbot, user_state]
     ).then(
-        lambda: None, None, image_inp
+        lambda: ("", None),
+        outputs=[text_inp, image_inp]
+    )
+    
+    clear_btn.click(
+        clear_chat,
+        outputs=[chatbot, image_inp]
     )
-
-    # Clear chat
-    def clear_chat():
-        return [], None
-    clear_btn.click(clear_chat, outputs=[chat, image_inp])
 
 if __name__ == "__main__":
-    demo.launch(server_name="0.0.0.0", server_port=7860, share=True)
+    # For Hugging Face Spaces, don't use share=True
+    try:
+        demo.launch(
+            server_name="0.0.0.0",
+            server_port=7860,
+            share=False,  # Changed to False for Hugging Face Spaces
+            show_error=True
+        )
+    except Exception as e:
+        print(f"Launch error: {e}")
+        # Fallback to simple launch
+        demo.launch(share=False)