Update src/streamlit_app.py

src/streamlit_app.py

@@ -1,4 +1,3 @@
-# streamlit_app.py
 import os
 import re
 import shutil
@@ -42,7 +41,6 @@ def clean_cache(max_size_gb: float = 2.0):
                     for f in files
                 )
                 size_gb = size_bytes / (1024 ** 3)
-                # remove if too big OR contains 'torch' to keep space free
                 if size_gb > max_size_gb or "torch" in folder:
                     shutil.rmtree(folder, ignore_errors=True)
                     total_deleted += size_gb
@@ -66,25 +64,19 @@ os.environ.update({
 })
 
 # -------------------------
-# App-specific imports (user modules)
+# App-specific imports
 # -------------------------
-# Make sure these modules exist (same as your earlier working project).
 from ingestion import extract_text_from_pdf, chunk_text
 from vectorstore import build_faiss_index
 from qa import retrieve_chunks, generate_answer, cache_embeddings, embed_chunks, genai_generate
 
 # -------------------------
-# Helper: TOC-based suggestion generator (robust)
+# Smart suggestion generator (fixed f-string)
 # -------------------------
-def generate_dynamic_suggestions_from_toc(toc: List[Tuple[str,str]], chunks: List[str], doc_name: str = "Document") -> List[str]:
-    """
-    Prefer TOC-derived natural suggestions. If genai_generate is available we can optionally use it,
-    otherwise produce deterministic, high-quality suggestions from TOC titles.
-    """
+def generate_dynamic_suggestions_from_toc(toc: List[Tuple[str, str]], chunks: List[str], doc_name="Document") -> List[str]:
     if not chunks:
         return []
 
-    # Clean TOC titles
     titles = []
     for sec, raw_title in toc or []:
         title = re.sub(r"^\s*[\dA-Za-z.\-]+\s*", "", raw_title)
@@ -93,55 +85,43 @@ def generate_dynamic_suggestions_from_toc(toc: List[Tuple[str,str]], chunks: Lis
             titles.append(title)
 
     if not titles:
-        # fallback to text-based suggestion: find verbs/imperatives in first chunks
         sample = " ".join(chunks[:4])
         sents = re.split(r'(?<=[.?!])\s+', sample)
         suggestions = []
         for s in sents:
-            if re.search(r'\b(set up|configure|how to|install|enable|prepare|step|procedure|process)\b', s, re.I):
-                clean = s.strip()
-                clean = re.sub(r'[.?!]+$', '', clean)
-                q = clean[0].upper() + clean[1:]
+            if re.search(r'\b(set up|configure|install|enable|procedure|process|how to|step)\b', s, re.I):
+                s = re.sub(r'[.?!]+$', '', s.strip())
+                q = s[0].upper() + s[1:]
                 if len(q) < 140:
                     suggestions.append(q if q.endswith('?') else q + '?')
-        return list(dict.fromkeys(suggestions))[:7]
+        return suggestions[:7]
 
-    # Try AI-based suggestion generation first (non-critical)
     try:
-        prompt = f"""You are an assistant that converts section titles into user-friendly, short questions.
+        prompt = f"""Generate 5-7 user-friendly, short questions based on the document's table of contents:
 Document: {doc_name}
-Table of contents:
-{chr(10).join(['- ' + t for t in titles[:20]])}
-Provide 5-7 concise, natural questions (each under 20 words) based on the TOC that a user might ask."""
+TOC:
+{chr(10).join(['- ' + t for t in titles[:15]])}"""
         ai_resp = genai_generate(prompt)
-        # extract lines that end with ?
         qs = re.findall(r'([^\n?.!]+\?)', ai_resp)
-        clean_qs = []
-        for q in qs:
-            q = q.strip()
-            if 8 < len(q) < 140:
-                clean_qs.append(q if q.endswith('?') else q + '?')
+        clean_qs = [q.strip() for q in qs if 8 < len(q) < 140]
         if clean_qs:
-            # dedupe preserving order
             return list(dict.fromkeys(clean_qs))[:7]
     except Exception:
-        # silently fallback to deterministic below
         pass
 
-    # Deterministic fallback: turn titles into "What is ..." or "How do I ..." where appropriate
     suggestions = []
     for t in titles[:15]:
         low = t.lower()
         if re.search(r'\b(set up|install|configure|enable|define|create|prepare)\b', low):
-            suggestions.append(f"How do I {re.sub(r'[^a-zA-Z0-9 \-]', '', low)}?")
+            cleaned = re.sub(r'[^a-zA-Z0-9 \-]', '', low)
+            suggestions.append(f"How do I {cleaned}?")
         elif re.search(r'\b(purpose|overview|objective|scope|what is)\b', low):
             suggestions.append(f"What is {t.strip().rstrip('.')}?")
         elif re.search(r'\b(step|procedure|process|task)\b', low):
             suggestions.append(f"What are the steps for {t.strip().rstrip('.')}?")
         else:
-            # generic
             suggestions.append(f"What is described in '{t}'?")
-    # dedupe & cleanup
+
     seen, final = set(), []
     for s in suggestions:
         s = re.sub(r'\s+', ' ', s).strip()
@@ -151,50 +131,29 @@ Provide 5-7 concise, natural questions (each under 20 words) based on the TOC th
     return final[:7]
 
 # -------------------------
-# CSS / UI polish
+# CSS for modern UI
 # -------------------------
-st.markdown(
-    """
-    <style>
-    /* layout tweaks */
-    .css-18e3th9 { padding-top: 2rem; } /* top padding */
-    .section-header { font-weight:700; font-size:1.2rem; margin-top:22px; margin-bottom:8px; color: #f3f4f6; }
-    .info-card { background: linear-gradient(180deg, rgba(16,24,39,0.9), rgba(6,10,14,0.9)); padding:12px 16px; border-radius:10px; color:#e6eef6; box-shadow: 0 6px 20px rgba(2,6,23,0.5); }
-    /* chips */
-    .suggest-chip {
-        display:inline-block; margin:6px 8px; padding:10px 16px; border-radius:999px;
-        background: rgba(31,41,55,0.65); border: 1px solid rgba(148,163,184,0.08);
-        color:#e6eef6; font-size:14px; cursor:pointer; transition: transform .12s ease, box-shadow .12s ease;
-    }
-    .suggest-chip:hover { transform: translateY(-4px); box-shadow: 0 8px 20px rgba(37,99,235,0.12); background: rgba(37,99,235,0.12); }
-    .suggest-chip.active { background: linear-gradient(90deg, rgba(59,130,246,0.12), rgba(99,102,241,0.06)); border:1px solid rgba(99,102,241,0.22); color:#eaf2ff; box-shadow: 0 8px 30px rgba(37,99,235,0.12); }
-    /* input */
-    .stTextInput>div>div>input { background: rgba(17,24,39,0.75); border-radius:8px; padding:12px 14px; color:#e6eef6; border:1px solid rgba(255,255,255,0.03); }
-    .stTextInput>div>div>input:focus { box-shadow: 0 6px 20px rgba(37,99,235,0.06); border:1px solid rgba(37,99,235,0.3); }
-    /* assistant card */
-    .assistant-card {
-        background: linear-gradient(180deg, rgba(6,10,14,0.7), rgba(10,15,20,0.6));
-        border-left: 4px solid rgba(59,130,246,0.9);
-        padding:18px 20px; border-radius:10px; color:#f8fbff; margin-bottom:10px;
-        box-shadow: 0 10px 40px rgba(2,6,23,0.5);
-    }
-    .assistant-note { color: #f1f5f9; margin-bottom:8px; }
-    .expander-half { background: rgba(255,255,255,0.02); border-radius:8px; padding:8px; margin-top:8px; }
-    /* compact section spacing */
-    .section-compact { margin-bottom: 10px; }
-    </style>
-    """,
-    unsafe_allow_html=True,
-)
+st.markdown("""
+<style>
+.section-header {font-weight:700;font-size:1.2rem;margin-top:22px;margin-bottom:8px;color:#f3f4f6;}
+.info-card {background:linear-gradient(180deg,rgba(16,24,39,0.9),rgba(6,10,14,0.9));padding:12px 16px;border-radius:10px;color:#e6eef6;box-shadow:0 6px 20px rgba(2,6,23,0.5);}
+.suggest-chip {display:inline-block;margin:6px 8px;padding:10px 16px;border-radius:999px;background:rgba(31,41,55,0.65);border:1px solid rgba(148,163,184,0.08);color:#e6eef6;font-size:14px;cursor:pointer;transition:transform .12s ease,box-shadow .12s ease;}
+.suggest-chip:hover {transform:translateY(-4px);box-shadow:0 8px 20px rgba(37,99,235,0.12);background:rgba(37,99,235,0.12);}
+.suggest-chip.active {background:linear-gradient(90deg,rgba(59,130,246,0.12),rgba(99,102,241,0.06));border:1px solid rgba(99,102,241,0.22);color:#eaf2ff;box-shadow:0 8px 30px rgba(37,99,235,0.12);}
+.stTextInput>div>div>input {background:rgba(17,24,39,0.75);border-radius:8px;padding:12px 14px;color:#e6eef6;border:1px solid rgba(255,255,255,0.03);}
+.stTextInput>div>div>input:focus {box-shadow:0 6px 20px rgba(37,99,235,0.06);border:1px solid rgba(37,99,235,0.3);}
+.assistant-card {background:linear-gradient(180deg,rgba(6,10,14,0.7),rgba(10,15,20,0.6));border-left:4px solid rgba(59,130,246,0.9);padding:18px 20px;border-radius:10px;color:#f8fbff;margin-bottom:10px;box-shadow:0 10px 40px rgba(2,6,23,0.5);}
+</style>
+""", unsafe_allow_html=True)
 
 # -------------------------
-# Page header
+# App Header
 # -------------------------
 st.title("📄 Enterprise Knowledge Assistant")
 st.caption("Query SAP documentation and enterprise PDFs — powered by reasoning and retrieval.")
 
 # -------------------------
-# Sidebar (compact)
+# Sidebar
 # -------------------------
 with st.sidebar:
     if "reasoning_mode" not in st.session_state:
@@ -212,36 +171,30 @@ with st.sidebar:
 
     st.markdown("---")
     st.header("⚙️ Settings")
-    chunk_size = st.slider("Chunk Size (chars)", min_value=200, max_value=1500, value=1000, step=50)
-    overlap = st.slider("Chunk Overlap (chars)", min_value=50, max_value=200, value=120, step=10)
-    top_k = st.slider("Top K Results", min_value=1, max_value=10, value=5)
+    chunk_size = st.slider("Chunk Size (chars)", 200, 1500, 1000, step=50)
+    overlap = st.slider("Chunk Overlap (chars)", 50, 200, 120, step=10)
+    top_k = st.slider("Top K Results", 1, 10, 5)
     st.markdown("---")
     st.caption("✨ Built by Shubham Sharma")
 
 # -------------------------
-# Initialize central session keys (safe)
+# Session state
 # -------------------------
 for k, v in {
     "show_more": False,
     "user_query_input": "",
-    "auto_run_after_suggest": False,
     "selected_suggestion_idx": None,
-    "last_doc_path": None,
-    "cached_embeddings_loaded": False,
 }.items():
     if k not in st.session_state:
         st.session_state[k] = v
 
 # -------------------------
-# File paths & base
+# Document processing
 # -------------------------
 BASE_DIR = os.path.dirname(__file__)
 SAMPLE_PATH = os.path.join(BASE_DIR, "sample.pdf")
 text, chunks, index, embeddings, toc = None, None, None, None, None
 
-# -------------------------
-# Document selection + upload handling
-# -------------------------
 if doc_choice == "-- Select --":
     st.info("⬅️ Please choose a document from the sidebar to begin.")
 else:
@@ -249,7 +202,7 @@ else:
         temp_path = SAMPLE_PATH
         st.success("📘 Using built-in Sample PDF.")
     else:
-        uploaded_file = st.file_uploader("📂 Upload your PDF (optional)", type="pdf")
+        uploaded_file = st.file_uploader("📂 Upload your PDF", type="pdf")
         if uploaded_file:
             temp_path = os.path.join("/tmp", uploaded_file.name)
             with open(temp_path, "wb") as f:
@@ -258,112 +211,54 @@ else:
         else:
             temp_path = None
 
-    # if user swapped to a different doc, clear previous session suggestions
-    if temp_path and st.session_state.get("last_doc_path") != temp_path:
-        st.session_state["last_doc_path"] = temp_path
-        st.session_state["user_query_input"] = ""
-        st.session_state["selected_suggestion_idx"] = None
-        st.session_state["auto_run_after_suggest"] = False
-        st.session_state["cached_embeddings_loaded"] = False
-
-    # process doc
     if temp_path:
-        with st.spinner("🔍 Extracting & processing document..."):
+        with st.spinner("🔍 Processing document..."):
             text, toc = extract_text_from_pdf(temp_path)
             chunks = chunk_text(text, chunk_size=chunk_size)
             st.markdown("<div class='info-card'>✅ Document loaded successfully.</div>", unsafe_allow_html=True)
 
-            # suggestions computed once per document (cached in session)
-            if "suggestions_for_doc" not in st.session_state or st.session_state.get("last_doc_path") != temp_path:
-                st.session_state["suggestions_for_doc"] = generate_dynamic_suggestions_from_toc(toc, chunks, os.path.basename(temp_path))
-
-        # embeddings + index (cache embeddings via provided helper)
-        with st.spinner("⚙️ Preparing embeddings & index..."):
-            try:
-                embeddings = cache_embeddings(os.path.basename(temp_path), chunks, embed_chunks)
-                st.session_state["cached_embeddings_loaded"] = True
-                st.success(f"📦 Loaded cached embeddings for '{os.path.basename(temp_path)}'.")
-            except Exception as e:
-                # still try to embed via embed_chunks
-                embeddings = embed_chunks(chunks)
-                st.session_state["cached_embeddings_loaded"] = False
+            query_suggestions = generate_dynamic_suggestions_from_toc(toc, chunks, os.path.basename(temp_path))
 
+        with st.spinner("⚙️ Preparing embeddings..."):
+            embeddings = cache_embeddings(os.path.basename(temp_path), chunks, embed_chunks)
             index = build_faiss_index(embeddings)
-            st.markdown("<div class='info-card'>🚀 Document ready — you can now ask questions below.</div>", unsafe_allow_html=True)
+            st.markdown("<div class='info-card'>🚀 Document ready — ask questions below.</div>", unsafe_allow_html=True)
 
         # -------------------------
-        # Ask a Question UI
+        # Ask a Question
         # -------------------------
-        st.markdown("<div class='section-header'>🤖 Ask a Question</div>", unsafe_allow_html=True)
+        st.markdown("## 🤖 Ask a Question")
 
-        suggestions = st.session_state.get("suggestions_for_doc") or []
-
-        # --- show top chips, show more toggle ---
-        if suggestions:
+        if query_suggestions:
             st.markdown("#### 💡 Suggested Questions")
-            visible = suggestions if st.session_state["show_more"] else suggestions[:3]
-
-            # lay out chips into up to 3 columns
-            cols = st.columns(3)
+            visible = query_suggestions if st.session_state.show_more else query_suggestions[:3]
+            cols = st.columns(min(3, len(visible)))
             for i, q in enumerate(visible):
-                # index offset in full suggestions
-                full_idx = i
                 col = cols[i % 3]
-                # chip button that sets the query
-                if col.button(f"🔍 {q}", key=f"sugg_btn_{full_idx}"):
-                    st.session_state["user_query_input"] = q
-                    st.session_state["selected_suggestion_idx"] = full_idx
-                    st.session_state["auto_run_after_suggest"] = True
-
-                # visual chip (for aesthetic)
-                chip_html = f"<div class='suggest-chip {'active' if st.session_state['selected_suggestion_idx']==full_idx else ''}'>{q}</div>"
-                col.markdown(chip_html, unsafe_allow_html=True)
-
-            toggle_text = "Show less ▲" if st.session_state["show_more"] else "Show more ▼"
-            if st.button(toggle_text, key="toggle_more"):
-                st.session_state["show_more"] = not st.session_state["show_more"]
+                if col.button(f"🔍 {q}", key=f"q_{i}"):
+                    st.session_state.user_query_input = q
+                    st.session_state.selected_suggestion_idx = i
+            toggle_text = "Show less ▲" if st.session_state.show_more else "Show more ▼"
+            if st.button(toggle_text):
+                st.session_state.show_more = not st.session_state.show_more
                 st.experimental_rerun()
 
-        # query input (canonical)
-        user_query = st.text_input("Type your question or pick one above:", value=st.session_state["user_query_input"], key="user_query_input")
+        user_query = st.text_input("Type your question or pick one above:", value=st.session_state.user_query_input)
 
-        # If a suggestion requested auto-run, run now (only once)
-        if st.session_state.get("auto_run_after_suggest"):
-            st.session_state["auto_run_after_suggest"] = False
-            # proceed to answer if query not empty (we fall through into the answer-rendering below)
-
-        # -------------------------
-        # Answer generation (single place)
-        # -------------------------
-        if user_query and user_query.strip():
+        if user_query.strip():
             st.caption("Mode: 🧠 Reasoning" if st.session_state.reasoning_mode else "Mode: 📄 Strict Document")
-            with st.spinner("💭 Retrieving context & generating answer..."):
+            with st.spinner("💭 Generating answer..."):
                 retrieved = retrieve_chunks(user_query, index, chunks, top_k=top_k, embeddings=embeddings)
                 answer = generate_answer(user_query, retrieved, reasoning_mode=st.session_state.reasoning_mode)
 
-            # assistant card (glassy)
-            st.markdown("<div class='section-header'>✅ Assistant’s Answer</div>", unsafe_allow_html=True)
-            st.markdown(f"<div class='assistant-card'><div class='assistant-note'>🧠 Assistant</div>{answer}</div>", unsafe_allow_html=True)
+            st.markdown("### ✅ Assistant’s Answer")
+            st.markdown(f"<div class='assistant-card'>{answer}</div>", unsafe_allow_html=True)
 
-            # supporting context + TOC collapsed (kept tidy & optional)
-            with st.expander("📄 Supporting Context", expanded=False):
+            with st.expander("📄 Supporting Context"):
                 for i, r in enumerate(retrieved, start=1):
-                    st.markdown(f"<div class='expander-half'><b>Chunk {i}:</b> {r}</div>", unsafe_allow_html=True)
-
-            if toc:
-                with st.expander("📚 Table of Contents (detected)", expanded=False):
-                    toc_text = "\n".join([f"{sec}. {title}" for sec, title in toc])
-                    st.text_area("TOC Preview", toc_text, height=160)
+                    st.markdown(f"**Chunk {i}:** {r}")
 
-        # -------------------------
-        # Document preview / stats
-        # -------------------------
         st.markdown("---")
         st.subheader("📑 Document Preview")
-        if text:
-            st.text_area("Extracted text (first 1000 chars)", text[:1000], height=200)
-            st.caption(f"📦 {len(chunks)} chunks processed.")
-        else:
-            st.info("No document text available yet.")
-
-# End of file
+        st.text_area("Extracted text (first 1000 chars)", text[:1000], height=200)
+        st.caption(f"📦 {len(chunks)} chunks processed.")