without comment

.streamlit/config.toml

@@ -1,5 +1,7 @@
-[server]
-headless = true
-
-[browser]
-gatherUsageStats = false
+[theme]
+base="light"
+primaryColor="#4A6D8C"
+backgroundColor="#ffffff"
+secondaryBackgroundColor="#f0f2f6"
+textColor="#31333F"
+font="sans serif"

DEPLOY_INSTRUCTIONS.md

@@ -1,45 +0,0 @@
-# Deploying to Hugging Face Spaces
-
-## Prerequisites
-- A Hugging Face account.
-- A Google API Key.
-
-## Steps
-
-1.  **Create a New Space**
-    - Go to [Hugging Face Spaces](https://huggingface.co/spaces).
-    - Click **"Create new Space"**.
-    - Enter a name (e.g., `academic-assistant`).
-    - **Select SDK**: Choose **Docker**. (This is critical because we need system-level dependencies like `graphviz` and specific Python versions).
-    - Choose "Public" or "Private".
-    - Click **"Create Space"**.
-
-2.  **Upload Files**
-    - You can upload files directly via the browser or use Git.
-    - **Files to Upload**:
-        - `Dockerfile`
-        - `requirements.txt`
-        - `app.py`
-        - `prompts.py`
-        - `core/` (The entire directory)
-        - `.streamlit/config.toml` (Optional, prevents welcome screen)
-    
-    *Note: Do NOT upload `.env` or your API keys directly in the files.*
-
-3.  **Configure Secrets (Environment Variables)**
-    - In your Space settings, go to the **"Settings"** tab.
-    - Scroll down to **"Variables and secrets"**.
-    - Click **"New secret"**.
-    - **Name**: `GOOGLE_API_KEY`
-    - **Value**: Your actual Google API Key (starting with `AIza...`).
-    - Click **Save**.
-
-4.  **Build & Run**
-    - Once files are uploaded, Hugging Face will automatically start building the Docker image.
-    - You can watch the "Build" logs.
-    - Once "Running", your app will be live!
-
-## Troubleshooting
-- **Build Error: "graphviz not found"**: Ensure your `Dockerfile` includes `RUN apt-get update && apt-get install -y graphviz`. (Already included).
-- **ModuleNotFoundError**: Ensure all packages (like `google-genai`) are in `requirements.txt`. (I just updated this for you).
-- **Runtime Error "Credentials"**: Ensure you added the `GOOGLE_API_KEY` in the Spaces Settings/Secrets, NOT just the `.env` file locally.

app.py

@@ -3,54 +3,105 @@ import os
 import time
 from dotenv import load_dotenv
 
-# Load env
 load_dotenv()
 
-from core.pdf_processer import PDFProcessor
+from core.retriever import Retriever
 from core.graph import RAGAgent
 from core.podcast import PodcastGenerator
 from core.visualizer import KnowledgeGraphGenerator
 from core.map_reduce import MapReduceSummarizer
 
-# Configuration
 st.set_page_config(
-    page_title="Academic Learning Assistant",
+    page_title="AI Knowledge Assistant",
     page_icon="🎓",
     layout="wide",
     initial_sidebar_state="collapsed"
 )
 
-# Custom CSS for "Fancy" Look
 st.markdown("""
 <style>
     .main { background-color: #f8f9fa; }
     
     /* Typography */
-    h1, h2, h3, h4 { font-family: 'Helvetica Neue', 'Inter', sans-serif; color: #333; }
+    h1, h2, h3, h4 { font-family: 'Helvetica Neue', 'Inter', sans-serif; color: #385A7C; }
+    p, li { color: #424242; line-height: 1.6; }
+    
+    /* Hero Section */
+    .hero-title { 
+        font-size: 3.5rem; 
+        font-weight: 800; 
+        color: #385A7C; 
+        text-align: center; 
+        margin-bottom: 0.5rem;
+        background: -webkit-linear-gradient(#4A6D8C, #385A7C);
+        -webkit-background-clip: text;
+        -webkit-text-fill-color: transparent;
+    }
+    .hero-subtitle { 
+        font-size: 1.5rem; 
+        color: #607d8b; 
+        text-align: center; 
+        margin-bottom: 2rem; 
+    }
+    
+    /* Section Headers */
+    .section-header {
+        font-size: 1.8rem;
+        font-weight: 700;
+        color: #385A7C;
+        margin-top: 2rem;
+        margin-bottom: 1rem;
+        border-left: 5px solid #4A6D8C;
+        padding-left: 15px;
+    }
     
     /* Card Style */
     .stCard {
-        background-color: white;
+        background-color: #ffffff;
         padding: 24px;
-        border-radius: 12px;
-        box-shadow: 0 4px 6px rgba(0,0,0,0.1);
+        border-radius: 16px;
+        box-shadow: 0 8px 20px rgba(56, 90, 124, 0.05);
         margin-bottom: 20px;
-        border: 1px solid #e1e4e8;
+        border: 1px solid #e1e8ed;
+        transition: transform 0.3s ease;
+    }
+    .stCard:hover {
+        transform: translateY(-5px);
+        box-shadow: 0 12px 30px rgba(56, 90, 124, 0.1);
     }
     
-    /* Button Styling */
-    .stButton>button {
+    /* Feature Badge */
+    .feature-badge {
+        background-color: #eef2f6;
+        color: #4A6D8C;
+        padding: 4px 12px;
         border-radius: 20px;
-        font-weight: 600;
+        font-size: 0.8rem;
+        font-weight: 700;
+        text-transform: uppercase;
+        margin-bottom: 10px;
+        display: inline-block;
     }
     
-    /* Navigation Button */
-    .nav-btn>button {
-        width: 100%;
-        background-color: #4CAF50;
-        color: white;
-        font-size: 18px;
-        padding: 12px;
+    /* Button Styling */
+    div.stButton > button {
+        border-radius: 30px !important;
+        padding: 10px 25px !important;
+        background-color: #4A6D8C !important;
+        color: white !important;
+        border: none !important;
+        box-shadow: 0 4px 12px rgba(74, 109, 140, 0.2) !important;
+        font-size: 1rem !important;
+        font-weight: 700 !important;
+    }
+    div.stButton > button:hover {
+        background-color: #385A7C !important;
+        color: white !important;
+        box-shadow: 0 6px 18px rgba(74, 109, 140, 0.3) !important;
+    }
+    /* Force white text for button labels */
+    div.stButton > button p {
+        color: white !important;
     }
 </style>
 """, unsafe_allow_html=True)
@@ -61,7 +112,7 @@ if "page" not in st.session_state:
 if "agent" not in st.session_state:
     st.session_state.agent = None
 if "pdf_processor" not in st.session_state:
-    st.session_state.pdf_processor = PDFProcessor()
+    st.session_state.pdf_processor = Retriever()
 if "messages" not in st.session_state:
     st.session_state.messages = []
 if "full_text" not in st.session_state:
@@ -81,13 +132,10 @@ def switch_page(page_name):
     st.session_state.page = page_name
     st.rerun()
 
-# --- PAGES ---
 def show_home():
-    # Hero Section
-    st.markdown("<h1 class='hero-title'>🎓 Academic Learning Assistant</h1>", unsafe_allow_html=True)
-    st.markdown("<p class='hero-subtitle'>Next-Gen Multi-Agent System for Research & Study</p>", unsafe_allow_html=True)
+    st.markdown("<h1 class='hero-title'>🎓 AI Knowledge Assistant</h1>", unsafe_allow_html=True)
+    st.markdown("<p class='hero-subtitle'>Transforming Complex Documents into Dynamic Multi-Modal Insights</p>", unsafe_allow_html=True)
     
-    # Call to Action Button (Centered)
     col_cta1, col_cta2, col_cta3 = st.columns([1, 1, 1])
     with col_cta2:
         if st.button("🚀 Launch Application", type="primary", width='stretch'):
@@ -95,18 +143,17 @@ def show_home():
     
     st.markdown("---")
 
-    # --- Section 1 & 2: Motivation & Problem (2-Column Layout) ---
     col1, col2 = st.columns(2, gap="large")
 
     with col1:
         st.markdown("<div class='section-header'>🌟 1. Motivation</div>", unsafe_allow_html=True)
         st.markdown("""
         <div class="stCard">
-            <p>In the era of information explosion, students and researchers are overwhelmed by the sheer volume of academic literature.</p>
+            <p>In the digital age, we are overwhelmed by the sheer volume of information and complex documents.</p>
             <ul>
-                <li><b>Information Overload:</b> Time-consuming to synthesize connections across lengthy PDFs.</li>
-                <li><b>Tool Limitations:</b> Basic search lacks semantics; generic LLMs hallucinate citations.</li>
-                <li><b>Multi-Modal Need:</b> Learners need text, audio (podcasts), and visual structures (graphs).</li>
+                <li><b>Information Overload:</b> Time-consuming to synthesize connections across lengthy files.</li>
+                <li><b>Tool Limitations:</b> Basic search lacks semantics; generic LLMs may lose context.</li>
+                <li><b>Multi-Modal Need:</b> Integrated text, audio (podcasts), and visual structures (graphs).</li>
             </ul>
         </div>
         """, unsafe_allow_html=True)
@@ -115,102 +162,103 @@ def show_home():
         st.markdown("<div class='section-header'>❓ 2. Problem Definition</div>", unsafe_allow_html=True)
         st.markdown("""
         <div class="stCard">
-            <p>Inefficiency and unreliability in synthesizing technical knowledge from unstructured documents.</p>
+            <p>Difficulty in reliably distilling and structuring insights from dense, unstructured data.</p>
             <ul>
-                <li><b>Reliability:</b> How to ensure answers are grounded and free from hallucinations?</li>
-                <li><b>Complexity:</b> How to grasp structural relationships in dense papers?</li>
-                <li><b>Accessibility:</b> How to make dry documents consumable on the go?</li>
+                <li><b>Reliability:</b> Ensuring answers are grounded in the source material.</li>
+                <li><b>Complexity:</b> Grasping underlying structural relationships in technical content.</li>
+                <li><b>Accessibility:</b> Making specialized documents consumable and structured.</li>
             </ul>
         </div>
         """, unsafe_allow_html=True)
 
-    # --- Section 3: Solution Approach (3-Column Feature Grid) ---
-    st.markdown("<div class='section-header'>💡 3. Solution Approach: Multi-Agent System</div>", unsafe_allow_html=True)
+    st.markdown("<div class='section-header'>💡 3. Versatile Multi-Agent Suite</div>", unsafe_allow_html=True)
     
-    c1, c2, c3 = st.columns(3, gap="medium")
+    c1, c2, c3, c4 = st.columns(4, gap="small")
     
     with c1:
         st.markdown("""
-        <div class="stCard">
-            <span class="feature-badge">The Researcher</span>
-            <h3>Reflective RAG Agent</h3>
-            <p>Utilizes a cyclic <b>Agentic Workflow</b>. It retrieves, answers, and performs a <b>Self-Reflection</b> step. If quality is poor, it automatically rewrites the query.</p>
+        <div class="stCard" style="min-height: 240px;">
+            <span class="feature-badge">Conversational</span>
+            <h4>Reflective RAG</h4>
+            <p style="font-size: 0.9rem;">A <b>LangGraph</b> state machine that retrieves and self-corrects via reasoning loops for grounded Q&A.</p>
         </div>
         """, unsafe_allow_html=True)
-        
+
     with c2:
         st.markdown("""
-        <div class="stCard">
-            <span class="feature-badge">The Producer</span>
-            <h3>Podcast Agent</h3>
-            <p>Employs <b>Map-Reduce</b> for long docs. Summarizes chunks, synthesizes a briefing, and uses <b>Gemini Native Audio</b> for natural multi-speaker dialogue.</p>
+        <div class="stCard" style="min-height: 240px;">
+            <span class="feature-badge">Synthesis</span>
+            <h4>Deep Summary</h4>
+            <p style="font-size: 0.9rem;">Utilizes <b>Map-Reduce</b> logic to distill long documents into high-density atomic facts and briefings.</p>
         </div>
         """, unsafe_allow_html=True)
         
     with c3:
         st.markdown("""
-        <div class="stCard">
-            <span class="feature-badge">The Analyst</span>
-            <h3>Visualizer Agent</h3>
-            <p>Analyzes text to extract entities and relationships, generating <b>Graphviz DOT</b> code to render interactive Knowledge Graphs for structural insight.</p>
+        <div class="stCard" style="min-height: 240px;">
+            <span class="feature-badge">Audio</span>
+            <h4>AI Podcast</h4>
+            <p style="font-size: 0.9rem;">Transforms facts into natural multi-speaker dialogue using <b>Gemini Native Audio</b> technology.</p>
+        </div>
+        """, unsafe_allow_html=True)
+        
+    with c4:
+        st.markdown("""
+        <div class="stCard" style="min-height: 240px;">
+            <span class="feature-badge">Visual</span>
+            <h4>Knowledge Graph</h4>
+            <p style="font-size: 0.9rem;">Maps relationships from summaries into hierarchical, interactive <b>DOT visuals</b> for structural insight.</p>
         </div>
         """, unsafe_allow_html=True)
 
-    # --- Section 4: System Architecture (Clean Layered View) ---
     st.markdown("<div class='section-header'>🏗️ 4. System Architecture</div>", unsafe_allow_html=True)
     
     st.markdown("""
     <div class="stCard">
         <div style="display: grid; grid-template-columns: repeat(4, 1fr); gap: 10px; text-align: center;">
             <div>
-                <h4>💻 Frontend</h4>
-                <p style="font-size: 0.9rem; color: #666;">Streamlit UI<br>Card Interface<br>Multi-file Upload</p>
+                <h4>🎨 Frontend</h4>
+                <p style="font-size: 0.85rem; color: #666;">Streamlit Dashboard<br>Responsive UI Components<br>Multi-modal Displays</p>
             </div>
             <div style="border-left: 1px solid #eee;">
-                <h4>🧠 Orchestration</h4>
-                <p style="font-size: 0.9rem; color: #666;">LangGraph<br>State Management<br>Cyclic Flows</p>
+                <h4>🧠 Brain</h4>
+                <p style="font-size: 0.85rem; color: #666;">LangChain / LangGraph<br>Agentic Workflows<br>Task Orchestration</p>
             </div>
             <div style="border-left: 1px solid #eee;">
-                <h4>💾 Data Layer</h4>
-                <p style="font-size: 0.9rem; color: #666;">ChromaDB<br>In-Memory Vector Store<br>HF Spaces Ready</p>
+                <h4>💾 Data</h4>
+                <p style="font-size: 0.85rem; color: #666;">ChromaDB Vector Store<br>Persistent Metadata<br>Hierarchical Retrieval</p>
             </div>
             <div style="border-left: 1px solid #eee;">
-                <h4>🤖 Model Layer</h4>
-                <p style="font-size: 0.9rem; color: #666;">Google Gemini 2.5<br>(Native Audio/Text)<br>NVIDIA NIM (Optional)</p>
+                <h4>🧬 Models</h4>
+                <p style="font-size: 0.85rem; color: #666;">NVIDIA Nemotron-3 (Reasoning)<br>Google Embedding-001 (Vector)<br>Gemini 2.5 Flash (Audio/TTS)</p>
             </div>
         </div>
     </div>
     """, unsafe_allow_html=True)
 
-    # --- Section 5: Implementation Details (Interactive Tabs) ---
     st.markdown("<div class='section-header'>⚙️ 5. Implementation Details</div>", unsafe_allow_html=True)
     
-    tab_rag, tab_pod, tab_vis = st.tabs(["🔄 RAG Loop", "🎙️ Map-Reduce Podcast", "🕸️ Knowledge Graph"])
+    tab_rag, tab_sum, tab_others = st.tabs(["💻Reflective RAG", "📄 Smart Summary", "🛠️ Tools & Visuals"])
     
     with tab_rag:
-        st.info("**Core Logic:** `core/graph.py`")
+        st.info("**Cyclic State Machine**")
         st.markdown("""
-        1. **Retrieve:** Fetches document chunks based on query.
-        2. **Generate:** Drafts answer with strict citations `[Source: paper.pdf]`.
-        3. **Reflect:** LLM acts as "grader" for relevance/hallucination.
-           - **Pass:** Returns final answer.
-           - **Fail:** Triggers **Query Rewrite** and re-enters Retrieve loop (Max 2 iters).
+        - Executes a reasoning loop: **Retrieve → Draft → Grade → Rewrite**.
+        - Powered by **LangGraph** to ensure answers are strictly evidence-based.
         """)
         
-    with tab_pod:
-        st.info("**Core Logic:** `core/podcast.py`")
+    with tab_sum:
+        st.info("**Map-Reduce Pipeline**")
         st.markdown("""
-        1. **Map:** Iterates through all PDF chunks to generate concise summaries.
-        2. **Reduce:** Aggregates summaries into a structured "Episode Briefing".
-        3. **Audio Synthesis:** Sends briefing to **Gemini 2.5 Flash** with prompt to "generate AUDIO directly" (Host & Expert personas).
+        - Seamlessly handles ultra-long documents by chunking and parallel summarizing.
+        - Provides the analytical foundation for deep-dive tools.
         """)
         
-    with tab_vis:
-        st.info("**Core Logic:** `core/visualizer.py`")
+    with tab_others:
+        st.info("**Multi-Modal Outputs**")
         st.markdown("""
-        1. **Extraction:** LLM extracts `Subject -> Predicate -> Object` triplets.
-        2. **Formatting:** Converts relationships into valid **Graphviz DOT** syntax.
-        3. **Rendering:** Streamlit renders the DOT code into a visual chart.
+        - **Podcast:** Multi-speaker audio briefings using Gemini TTS.
+        - **Knowledge Graph:** Structural relationship mapping via DOT syntax.
         """)
 
     st.markdown("<br><br>", unsafe_allow_html=True)
@@ -221,18 +269,16 @@ def ensure_deep_summary():
         
     if not st.session_state.deep_summary:
         if st.session_state.full_text:
-            with st.spinner("Analyzing Document (Map-Reduce Deep Summary)..."):
+            with st.spinner("Analyzing Document (Deep Summary)..."):
                 mr = MapReduceSummarizer()
                 st.session_state.deep_summary = mr.generate_deep_summary(st.session_state.full_text)
     return st.session_state.deep_summary
 
-# Dialog Helper (handles version differences)
 if hasattr(st, "dialog"):
     dialog_decorator = st.dialog
 elif hasattr(st, "experimental_dialog"):
     dialog_decorator = st.experimental_dialog
 else:
-    # No dialog support, fallback to a simple function
     def dialog_decorator(*args, **kwargs):
         def decorator(func):
             return func
@@ -247,7 +293,6 @@ def view_summary_dialog(text):
 @dialog_decorator("Knowledge Graph Visualization", width="large")
 def view_graph_dialog(dot_code):
     st.graphviz_chart(dot_code, width="stretch")
-    st.caption("Right-click -> 'Open Image in New Tab' to zoom/download.")
 
 def show_app():
     # Sidebar: Clean, just for upload and nav
@@ -275,7 +320,6 @@ def show_app():
                     for f in new_files:
                         st.session_state.processed_files.add(f.name)
                     
-                    # Set success message and reset uploader to clear the list
                     st.session_state.upload_status = f"Successfully indexed ~{total_tokens:,} tokens from {len(new_files)} new file(s)."
                     st.session_state.uploader_key += 1
                     st.rerun()
@@ -284,7 +328,7 @@ def show_app():
             st.success("Analysis Ready")
             
             if st.button("🔄 Reset / Clear All", type="primary"):
-                st.session_state.pdf_processor = PDFProcessor()
+                st.session_state.pdf_processor = Retriever()
                 st.session_state.agent = None
                 st.session_state.messages = []
                 st.session_state.full_text = ""
@@ -295,13 +339,11 @@ def show_app():
                 st.rerun()
 
 
-    # Layout: Chat (Left/Center) | Tools (Right)
     col_chat, col_tools = st.columns([3, 1.3])
 
     with col_chat:
         st.subheader("💬 Chat")
         
-        # Chat History
         for msg in st.session_state.messages:
             with st.chat_message(msg["role"]):
                 if "thoughts" in msg and msg["thoughts"]:
@@ -310,7 +352,6 @@ def show_app():
                             st.write(log)
                 st.markdown(msg["content"])
         
-        # User Input
         if prompt := st.chat_input("Ask about the document..."):
             st.session_state.messages.append({"role": "user", "content": prompt})
             with st.chat_message("user"):
@@ -318,7 +359,6 @@ def show_app():
             
             with st.chat_message("assistant"):
                 if st.session_state.agent:
-                    # Container for intermediate thought process
                     with st.status("Agent Reasoning...", expanded=True) as status:
                         thoughts = []
                         
@@ -346,7 +386,6 @@ def show_app():
                         
                         response = result["generation"]
                         
-                        # Show debug steps comfortably (Optional redundant info, maybe keep for final stats)
                         with st.expander("📊 Final Stats", expanded=False):
                             st.write(f"**Reflected:** {result.get('reflection_score')} | **Total Iter:** {result.get('iterations')}")
                         
@@ -378,7 +417,6 @@ def show_app():
                          st.session_state.deep_summary = None
                          st.rerun()
             
-            # Podcast Tool
             with st.expander("🎧 Podcast", expanded=False):
                 if not st.session_state.podcast_audio:
                     if st.button("Generate Audio"):
@@ -399,20 +437,18 @@ def show_app():
                          st.session_state.podcast_audio = None
                          st.rerun()
 
-            # Knowledge Graph Tool
             with st.expander("🕸️ Knowledge Graph", expanded=False):
                 if not st.session_state.graph_dot:
                     if st.button("Generate Graph"):
                         summary_text = ensure_deep_summary()
                         with st.spinner("Building Graph structure..."):
                             kg_gen = KnowledgeGraphGenerator()
-                            # Generate and clean dot code
                             raw_dot = kg_gen.generate_graph(summary_text)
                             st.session_state.graph_dot = raw_dot
                             st.rerun()
                 else:
                     st.success("Graph Ready!")
-                    if st.button("👁️ View Graph High-Res", type="primary", width='stretch'):
+                    if st.button("👁️ View Knowledge Graph", type="primary", width='stretch'):
                          view_graph_dialog(st.session_state.graph_dot)
                     
                     if st.button("🔄 Regenerate Graph"):

core/graph.py

@@ -28,7 +28,6 @@ class RAGAgent:
         question = state["question"]
         docs = state["documents"]
         
-        # Format context with explicit source numbering for citation
         context = "\n\n".join([f"[Document: {doc.metadata.get('filename', 'Unknown')} | Page: {doc.metadata.get('page', 0) + 2}] {doc.page_content}" for doc in docs])
         
         chain = RAG_PROMPT | self.llm | StrOutputParser()
@@ -40,7 +39,6 @@ class RAGAgent:
         generation = state["generation"]
         docs = state["documents"]
         
-        # Format context so the reflector can check for grounding
         context = "\n\n".join([f"[Source: {doc.metadata.get('filename', 'Unknown')}] {doc.page_content}" for doc in docs])
         
         chain = REFLECTION_PROMPT | self.llm | StrOutputParser()
@@ -50,7 +48,6 @@ class RAGAgent:
             "generation": generation
         })
         
-        # Normalize score
         normalized_score = "yes" if "yes" in score.lower() else "no"
         return {"reflection_score": normalized_score}
 
@@ -80,13 +77,12 @@ class RAGAgent:
     def build_graph(self):
         workflow = StateGraph(GraphState)
 
-        # Define Nodes
+
         workflow.add_node("retrieve", self.retrieve)
         workflow.add_node("generate", self.generate)
         workflow.add_node("reflect", self.reflect)
         workflow.add_node("rewrite_query", self.rewrite_query)
 
-        # Build Edges
         workflow.set_entry_point("retrieve")
         workflow.add_edge("retrieve", "generate")
         workflow.add_edge("generate", "reflect")
@@ -112,21 +108,15 @@ class RAGAgent:
         }
         
         final_state = inputs
-        # Stream allowing for intermediate updates
         for output in self.app.stream(inputs):
             for key, value in output.items():
-                # Update our tracking of final state (simplistic merge)
                 final_state.update(value)
-                
-                # 'key' is the node name (e.g., 'retrieve', 'generate')
-                # 'final_state' is the cumulative state
                 if callback:
                     callback(key, final_state)
                 
         return final_state
 
     def get_graph_image(self, file_path: str = None):
-        """Returns the PNG binary of the graph structure. Optionally saves to file."""
         img_bytes = self.app.get_graph().draw_mermaid_png()
         if file_path:
             with open(file_path, "wb") as f:

core/models.py

@@ -5,12 +5,7 @@ from langchain_google_genai import GoogleGenerativeAIEmbeddings, ChatGoogleGener
 from google import genai
 
 def get_llm(model_name: str = "nvidia/nemotron-3-nano-30b-a3b"):
-    """
-    Returns a configured ChatGoogleGenerativeAI instance.
-    Uses st.secrets or environment variable for API Key.
-    """
     api_key = os.getenv("NV_API_KEY")
-    # Streamlit Cloud deployment support
     if not api_key and "NV_API_KEY" in st.secrets:
         api_key = st.secrets["NV_API_KEY"]
     
@@ -27,9 +22,6 @@ def get_llm(model_name: str = "nvidia/nemotron-3-nano-30b-a3b"):
     )
 
 def get_embeddings():
-    """
-    Returns GoogleGenerativeAIEmbeddings.
-    """
     api_key = os.getenv("GOOGLE_API_KEY")
     if not api_key and "GOOGLE_API_KEY" in st.secrets:
         api_key = st.secrets["GOOGLE_API_KEY"]
@@ -43,10 +35,6 @@ from google.genai import types
 
 
 def generate_podcast_audio(script_text: str):
-    """
-    Calls Gemini TTS with multi-speaker configuration.
-    Returns raw audio data.
-    """
     api_key = os.getenv("GOOGLE_API_KEY")
     if not api_key and "GOOGLE_API_KEY" in st.secrets:
         api_key = st.secrets["GOOGLE_API_KEY"]

core/podcast.py

@@ -20,15 +20,10 @@ class PodcastGenerator:
         return chain.invoke({"text": briefing_text})
 
     def generate_audio_file(self, script_text):
-        """
-        Uses centralized Gemini TTS logic.
-        """
         try:
             data = generate_podcast_audio(script_text)
             
             if data:
-                # Use NamedTemporaryFile to get a unique name, then close it immediately
-                # so wave.open can re-open it for writing.
                 with tempfile.NamedTemporaryFile(delete=False, suffix=".wav") as tmp:
                     tmp_path = tmp.name
                 
@@ -41,7 +36,6 @@ class PodcastGenerator:
             print(f"Gemini Official TTS failed: {e}. Falling back to gTTS.")
             try:
                 from gtts import gTTS
-                # Strip names for gTTS single voice
                 tts_text = script_text.replace("Alex:", "").replace("Jamie:", "")
                 tts = gTTS(tts_text, lang='en')
                 with tempfile.NamedTemporaryFile(delete=False, suffix=".mp3") as fp:

core/{pdf_processer.py → retriever.py}

@@ -7,7 +7,7 @@ import tempfile
 
 import tiktoken
 
-class PDFProcessor:
+class Retriever:
     def __init__(self):
         self.embeddings = get_embeddings()
         self.vector_store = None
@@ -18,14 +18,9 @@ class PDFProcessor:
         return len(self.encoding.encode(text))
 
     def process_pdf(self, uploaded_files):
-        """
-        Reads uploaded PDF file(s), splits text, and initializes VectorStore.
-        Returns the total token count of newly added text.
-        """
         if not uploaded_files:
             return 0
             
-        # Ensure input is a list (handle single file case just in case)
         if not isinstance(uploaded_files, list):
             uploaded_files = [uploaded_files]
 
@@ -33,7 +28,6 @@ class PDFProcessor:
         cumulative_text = ""
 
         for uploaded_file in uploaded_files:
-            # Save uploaded file typically to temp because PyPDFLoader needs a path
             with tempfile.NamedTemporaryFile(delete=False, suffix=".pdf") as tmp_file:
                 tmp_file.write(uploaded_file.read())
                 tmp_path = tmp_file.name
@@ -61,12 +55,11 @@ class PDFProcessor:
         )
         splits = text_splitter.split_documents(new_documents)
         
-        # Initialize Chroma (In-memory)
         if self.vector_store is None:
             self.vector_store = Chroma.from_documents(
                 documents=splits,
                 embedding=self.embeddings,
-                collection_name="academic_papers"
+                collection_name="knowledge_base"
             )
         else:
             self.vector_store.add_documents(splits)

core/visualizer.py

@@ -1,8 +1,5 @@
 from core.models import get_llm
-from langchain_core.prompts import PromptTemplate
 from langchain_core.output_parsers import StrOutputParser
-import graphviz
-import re
 from prompts import GRAPH_PROMPT
 
 class KnowledgeGraphGenerator:
@@ -10,15 +7,12 @@ class KnowledgeGraphGenerator:
         self.llm = get_llm()
 
     def generate_graph(self, text):
-        # Text is now the "Deep Summary", so no need to truncate.
-        input_text = text 
-        
         chain = GRAPH_PROMPT | self.llm | StrOutputParser()
-        dot_code = chain.invoke({"text": input_text})
         
-        # Cleanup markdown if present
+        dot_code = chain.invoke({"text": text})
+       
         dot_code = dot_code.replace("```dot", "").replace("```", "").strip()
         if "digraph" not in dot_code:
             dot_code = f'digraph G {{ {dot_code} }}'
-            
+        
         return dot_code

prompts.py

@@ -1,11 +1,11 @@
 from langchain_core.prompts import ChatPromptTemplate
 
 # RAG Generation Prompt
-RAG_SYSTEM = """You are a research assistant. Answer the user's question based strictly on the provided context. 
-If the context does not contain the answer, say "I cannot answer this based on the document."
+RAG_SYSTEM = """You are a professional research and learning assistant. Answer the user's question based strictly on the provided context. 
+If the context does not contain the answer, say "I cannot answer this based on the provided material."
 
 Requirements:
-1. Use academic tone.
+1. Use a professional and objective tone.
 2. **In-text Citations:** Use Unicode Superscript Numbers (¹, ², ³, ⁴, ⁵, ⁶, ⁷, ⁸, ⁹, ¹⁰) strictly. Place them immediately after the punctuation or relevant phrase.
    - Do NOT use `[^1]` (Markdown footnotes) or `[1]` (Brackets). 
    - Example: ...at compile time¹.
@@ -135,7 +135,7 @@ PODCAST_AUDIO_PROMPT = ChatPromptTemplate.from_messages([
 
 
 # Knowledge Graph Prompt
-GRAPH_SYSTEM = """You are an expert at visualizing complex knowledge information. 
+GRAPH_SYSTEM = """You are an expert at visualizing complex information. 
 Your goal is to extract a DEEP hierarchical structure and key relationships from the provided text and represent them as a CLEAN, multi-level Knowledge Graph using DOT syntax.
 
 CRITICAL INSTRUCTIONS: