Update app.py

app.py

@@ -4,139 +4,47 @@ import fitz
 import re
 import numpy as np
 import faiss
-import os
 from sentence_transformers import SentenceTransformer
 from groq import Groq
+from faster_whisper import WhisperModel
 
 # =========================
 # INITIALIZE MODELS
 # =========================
 
 embedding_model = SentenceTransformer("all-MiniLM-L6-v2")
+whisper_model = WhisperModel("base", compute_type="int8")
 
-whisper_model = None
+# 🔑 PUT YOUR GROQ API KEY HERE
+client = Groq(api_key="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")
 
-def get_whisper():
-    global whisper_model
-    if whisper_model is None:
-        from faster_whisper import WhisperModel
-        whisper_model = WhisperModel("base", compute_type="int8")
-    return whisper_model
-
-GROQ_API_KEY = os.environ.get("YOUR_GROQ_API_KEY", "")
-if not GROQ_API_KEY:
-    raise ValueError("YOUR_GROQ_API_KEY secret is not set. Add it in HF Space Settings → Secrets.")
-
-client = Groq(api_key=GROQ_API_KEY)
+# Use stable model
 MODEL_NAME = "llama-3.3-70b-versatile"
 
+# Global storage
 sections = {}
 section_texts = []
 index = None
 
+
 # =========================
 # PDF FUNCTIONS
 # =========================
 
-def doi_to_pdf_url(doi):
-    """
-    Given a DOI, try multiple strategies to find a downloadable PDF:
-    1. Unpaywall API  — finds legal open-access PDFs for any DOI (no key needed)
-    2. Direct arXiv   — if the DOI belongs to an arXiv paper
-    3. Europe PMC     — broad biomedical / life-science coverage
-    Returns (pdf_url, paper_title).
-    """
-    doi = doi.strip()
-    # Strip common DOI prefixes so bare DOI always works
-    for prefix in ("https://doi.org/", "http://doi.org/", "doi:", "DOI:"):
-        if doi.startswith(prefix):
-            doi = doi[len(prefix):]
-            break
-
-    title = None
-
-    # ── Strategy 1: Unpaywall (free, ~85% OA coverage) ──────────────
+def download_arxiv_pdf(arxiv_id):
     try:
-        email = "research@assistant.app"
-        r = requests.get(
-            f"https://api.unpaywall.org/v2/{doi}?email={email}", timeout=15
-        )
-        if r.status_code == 200:
-            data = r.json()
-            title = data.get("title")
-            best = data.get("best_oa_location")
-            if best:
-                pdf_url = best.get("url_for_pdf") or best.get("url")
-                if pdf_url:
-                    return pdf_url, title
-            for loc in data.get("oa_locations", []):
-                pdf_url = loc.get("url_for_pdf") or loc.get("url")
-                if pdf_url:
-                    return pdf_url, title
-    except Exception as e:
-        print(f"Unpaywall error: {e}")
-
-    # ── Strategy 2: arXiv DOI pattern ───────────────────────────────
-    try:
-        arxiv_match = re.search(r"arXiv[\.:](\d{4}\.\d{4,5})", doi, re.IGNORECASE)
-        if arxiv_match:
-            arxiv_id = arxiv_match.group(1)
-            return f"https://arxiv.org/pdf/{arxiv_id}.pdf", title
-    except Exception as e:
-        print(f"arXiv DOI parse error: {e}")
-
-    # ── Strategy 3: Europe PMC ───────────────────────────────────────
-    try:
-        r = requests.get(
-            f"https://www.ebi.ac.uk/europepmc/webservices/rest/search"
-            f"?query=DOI:{doi}&format=json&resultType=core",
-            timeout=15,
-        )
-        if r.status_code == 200:
-            results = r.json().get("resultList", {}).get("result", [])
-            if results:
-                item = results[0]
-                title = title or item.get("title")
-                pmcid = item.get("pmcid")
-                if pmcid:
-                    pdf_url = (
-                        f"https://europepmc.org/backend/ptpmcrender.fcgi"
-                        f"?accid={pmcid}&blobtype=pdf"
-                    )
-                    return pdf_url, title
-    except Exception as e:
-        print(f"Europe PMC error: {e}")
-
-    return None, title
-
-
-def download_pdf_from_doi(doi):
-    """Resolve DOI → PDF URL → download to /tmp. Returns (file_path, paper_title)."""
-    try:
-        pdf_url, title = doi_to_pdf_url(doi)
-        if not pdf_url:
-            return None, title
-
-        safe_name = re.sub(r"[^\w\-]", "_", doi)[:60]
-        file_path = f"/tmp/{safe_name}.pdf"
-
-        headers = {"User-Agent": "Mozilla/5.0 (ResearchAssistant/1.0)"}
-        r = requests.get(pdf_url, timeout=40, headers=headers, allow_redirects=True)
-        r.raise_for_status()
-
-        # Verify it's actually a PDF
-        if b"%PDF" not in r.content[:16]:
-            print(f"Response is not a PDF from {pdf_url}")
-            return None, title
+        url = f"https://arxiv.org/pdf/{arxiv_id}.pdf"
+        response = requests.get(url)
+        response.raise_for_status()
 
+        file_path = f"{arxiv_id}.pdf"
         with open(file_path, "wb") as f:
-            f.write(r.content)
+            f.write(response.content)
 
-        return file_path, title
+        return file_path
+    except:
+        return None
 
-    except Exception as e:
-        print(f"PDF download error: {e}")
-        return None, None
 
 def extract_text_from_pdf(pdf_path):
     doc = fitz.open(pdf_path)
@@ -145,641 +53,247 @@ def extract_text_from_pdf(pdf_path):
         text += page.get_text()
     return text
 
+
+# =========================
+# ROBUST SECTION EXTRACTION
+# =========================
+
 def extract_sections(text):
+
     patterns = [
-        r"\n([IVX]+\.\s+[A-Z][A-Z\s]+)",
-        r"\n(\d+\.\d+\.\d+\s+[^\n]+)",
-        r"\n(\d+\.\d+\s+[^\n]+)",
-        r"\n(\d+\.\s+[^\n]+)",
-        r"\n(\d+\s+[^\n]+)",
-        r"\n([A-Z][A-Z\s]{4,})\n"
+        r"\n([IVX]+\.\s+[A-Z][A-Z\s]+)",   # Roman
+        r"\n(\d+\.\d+\.\d+\s+[^\n]+)",     # 1.1.1
+        r"\n(\d+\.\d+\s+[^\n]+)",          # 1.1
+        r"\n(\d+\.\s+[^\n]+)",             # 1.
+        r"\n(\d+\s+[^\n]+)",               # 1
+        r"\n([A-Z][A-Z\s]{4,})\n"          # ALL CAPS
     ]
+
     matches = []
     for p in patterns:
         matches.extend(list(re.finditer(p, text)))
+
     matches = sorted(matches, key=lambda x: x.start())
+
     extracted = {}
+
     for i, match in enumerate(matches):
         title = match.group(1).strip()
+
         start = match.end()
-        end = matches[i + 1].start() if i + 1 < len(matches) else len(text)
+        end = matches[i+1].start() if i+1 < len(matches) else len(text)
+
         content = text[start:end].strip()
+
         if len(content) > 4000:
             content = content[:4000]
+
         extracted[title] = content
+
+    # Add abstract manually
     abstract_match = re.search(r"Abstract(.*?)\n", text, re.DOTALL)
     if abstract_match:
         extracted["Abstract"] = abstract_match.group(1).strip()
+
     return extracted
 
+
+# =========================
+# VECTOR STORE
+# =========================
+
 def build_vector_store(sections_dict):
     global index, section_texts
+
     section_texts = list(sections_dict.values())
+
     if len(section_texts) == 0:
         index = None
         return
+
     embeddings = embedding_model.encode(section_texts)
     embeddings = np.array(embeddings).astype("float32")
+
     dim = embeddings.shape[1]
     index = faiss.IndexFlatL2(dim)
     index.add(embeddings)
 
-def load_paper(doi):
+
+# =========================
+# LOAD PAPER
+# =========================
+
+def load_paper(arxiv_id):
     global sections
-    if not doi or not doi.strip():
-        return gr.update(choices=[]), "⚠️ Please enter a valid DOI"
-    pdf_path, title = download_pdf_from_doi(doi)
+
+    pdf_path = download_arxiv_pdf(arxiv_id)
+
     if pdf_path is None:
-        return gr.update(choices=[]), (
-            "❌ Could not find or download a PDF for this DOI.\n\n"
-            "**Tips:** Make sure the paper is open-access. "
-            "Try formats like `10.1038/s41586-021-03819-2` or paste the full `https://doi.org/...` URL."
-        )
+        return gr.update(choices=[]), "❌ Invalid arXiv ID"
+
     text = extract_text_from_pdf(pdf_path)
     sections = extract_sections(text)
-    if not sections:
-        return gr.update(choices=[]), "⚠️ PDF downloaded but no sections could be detected."
+
     build_vector_store(sections)
-    label = f"✅ **{title}**\n\n{len(sections)} sections indexed and ready" if title else f"✅ Paper loaded — {len(sections)} sections indexed and ready"
-    return gr.update(choices=list(sections.keys())), label
+
+    return gr.update(choices=list(sections.keys())), "✅ Paper Loaded Successfully"
+
+
+# =========================
+# SUMMARY FUNCTION
+# =========================
 
 def summarize_section(section_title):
     try:
         if not sections:
-            return "⚠️ Please load a paper first."
-        if not section_title or section_title not in sections:
-            return "⚠️ Please select a section from the dropdown."
-        content = sections[section_title][:4000]
+            return "❌ Load paper first"
+
+        if section_title not in sections:
+            return "❌ Section not found"
+
+        content = sections[section_title]
+
         if not content:
-            return "⚠️ This section appears to be empty."
-        prompt = f"""You are a research assistant. Summarize the following section from a research paper in a clear, structured way.
+            return "❌ Empty section"
 
-Provide:
-• **Main Idea** — What is this section about?
-• **Key Concepts** — What are the important terms or methods?
-• **Simple Explanation** — Explain it as if to a graduate student unfamiliar with the topic
-• **Why It Matters** — What is the contribution or significance?
+        content = content[:4000]
 
-Section Title: {section_title}
+        prompt = f"""
+Summarize this research section:
+
+- Main idea
+- Key concepts
+- Simple explanation
+- Importance
+
+Section: {section_title}
 
 Content:
 {content}
 """
+
         response = client.chat.completions.create(
             model=MODEL_NAME,
             messages=[{"role": "user", "content": prompt}],
             temperature=0.3
         )
+
         return response.choices[0].message.content
+
     except Exception as e:
-        return f"❌ Error: {str(e)}"
+        return f"❌ Error:\n{str(e)}"
+
+
+# =========================
+# RAG CHAT
+# =========================
 
 def rag_chat(message, history):
     try:
         global index
-        if not message or not message.strip():
-            return history, ""
+
         if index is None:
-            history = history + [[None, "⚠️ Please load a paper first."]]
+            history.append({"role": "assistant", "content": "❌ Load paper first"})
             return history, ""
+
         query_embedding = embedding_model.encode([message])
         query_embedding = np.array(query_embedding).astype("float32")
+
         D, I = index.search(query_embedding, k=3)
-        retrieved = "\n\n".join([section_texts[i] for i in I[0] if i < len(section_texts)])
-        prompt = f"""You are an expert research assistant. Answer the question using ONLY the provided context from the paper.
-If the answer cannot be found in the context, say so clearly. Be precise and cite relevant details.
+
+        retrieved = "\n\n".join([section_texts[i] for i in I[0]])
+
+        prompt = f"""
+Answer using ONLY this context.
 
 Context:
 {retrieved}
 
-Question: {message}
+Question:
+{message}
 """
+
         response = client.chat.completions.create(
             model=MODEL_NAME,
             messages=[{"role": "user", "content": prompt}],
             temperature=0.2
         )
+
         answer = response.choices[0].message.content
-        history = history + [[message, answer]]
+
+        # ✅ FIXED FORMAT
+        history.append({"role": "user", "content": message})
+        history.append({"role": "assistant", "content": answer})
+
         return history, ""
+
     except Exception as e:
-        history = history + [[message, f"❌ Error: {str(e)}"]]
+        history.append({"role": "assistant", "content": f"❌ Error:\n{str(e)}"})
         return history, ""
 
+
+# =========================
+# VOICE CHAT
+# =========================
+
 def voice_chat(audio, history):
     try:
         if audio is None:
-            history = history + [[None, "⚠️ No audio received. Please record a voice message."]]
-            return history, ""
-        model = get_whisper()
-        segments, _ = model.transcribe(audio)
-        text = " ".join([seg.text for seg in segments]).strip()
-        if not text:
-            history = history + [[None, "⚠️ Could not transcribe audio. Please try again."]]
             return history, ""
+
+        segments, _ = whisper_model.transcribe(audio)
+        text = " ".join([seg.text for seg in segments])
+
         return rag_chat(text, history)
+
     except Exception as e:
-        history = history + [[None, f"❌ Error: {str(e)}"]]
+        history.append({"role": "assistant", "content": f"❌ Error:\n{str(e)}"})
         return history, ""
 
+
 # =========================
-# CSS — "Obsidian Intelligence" Premium 2026 Research Platform
-# Ethos: High-Density Functionalism meets Dark Luxury
-# Trends: Bento Grid · Glassmorphism · Kinetic borders · Layered depth
+# UI
 # =========================
 
-CSS = """
-@import url('https://fonts.googleapis.com/css2?family=Instrument+Serif:ital@0;1&family=Geist:wght@300;400;500;600;700&family=Geist+Mono:wght@400;500&display=swap');
-
-/* ═══════════ DESIGN TOKENS ═══════════ */
-:root {
-    --void:       #07080c;
-    --bg-0:       #0b0d13;
-    --bg-1:       #0f1119;
-    --bg-2:       #13161f;
-    --bg-3:       #181c28;
-    --bg-4:       #1e2333;
-    --bg-5:       #242840;
-    --b-0:        #1c2030;
-    --b-1:        #252a3c;
-    --b-2:        #303650;
-    --b-hi:       #404868;
-    --gold:       #d4a853;
-    --gold-lt:    #e8c47a;
-    --gold-dk:    #9e7535;
-    --gold-muted: rgba(212,168,83,0.08);
-    --gold-glow:  rgba(212,168,83,0.16);
-    --gold-ring:  0 0 0 2.5px rgba(212,168,83,0.22);
-    --green:      #3ecf8e;
-    --blue:       #4d94ff;
-    --red:        #f06565;
-    --amber:      #f0a443;
-    --tx-1:       #f0ece2;
-    --tx-2:       #8a8578;
-    --tx-3:       #4a4740;
-    --r-xs: 6px; --r-sm: 10px; --r: 16px; --r-lg: 22px; --r-xl: 28px;
-    --sh-sm:  0 2px 12px rgba(0,0,0,.4);
-    --sh:     0 4px 32px rgba(0,0,0,.55);
-    --sh-lg:  0 8px 64px rgba(0,0,0,.7);
-    --sh-xl:  0 16px 80px rgba(0,0,0,.8);
-    --sh-gold: 0 4px 28px rgba(212,168,83,0.2);
-}
-
-*, *::before, *::after { box-sizing: border-box; margin: 0; }
-html { scroll-behavior: smooth; }
-
-body, .gradio-container {
-    background: var(--bg-0) !important;
-    font-family: 'Geist', system-ui, sans-serif !important;
-    color: var(--tx-1) !important;
-    min-height: 100vh;
-    -webkit-font-smoothing: antialiased;
-}
-
-/* Dot-matrix background */
-.gradio-container {
-    background-image: radial-gradient(circle, rgba(212,168,83,0.05) 1px, transparent 1px) !important;
-    background-size: 28px 28px !important;
-    background-attachment: fixed !important;
-}
-
-/* Vignette overlay */
-.gradio-container::before {
-    content: '';
-    position: fixed; inset: 0;
-    background: radial-gradient(ellipse 80% 60% at 50% 0%, transparent 40%, rgba(7,8,12,0.75) 100%);
-    pointer-events: none; z-index: 0;
-}
-
-/* ═══════════ HERO HEADER ═══════════ */
-#hero {
-    position: relative;
-    background: linear-gradient(145deg, #0f1220 0%, #131828 45%, #0c0f1a 100%);
-    border: 1px solid var(--b-2);
-    border-radius: var(--r-xl);
-    padding: 56px 48px 44px;
-    text-align: center;
-    overflow: hidden;
-    box-shadow: var(--sh-xl);
-    margin-bottom: 2px;
-}
-
-#hero::before {
-    content: '';
-    position: absolute;
-    top: -120px; left: 50%; transform: translateX(-50%);
-    width: 600px; height: 400px;
-    background: conic-gradient(from 180deg at 50% 50%, rgba(212,168,83,0) 0deg, rgba(212,168,83,0.07) 60deg, rgba(77,148,255,0.04) 120deg, rgba(212,168,83,0) 180deg, rgba(212,168,83,0.06) 240deg, rgba(77,148,255,0.03) 300deg, rgba(212,168,83,0) 360deg);
-    animation: aurora 12s linear infinite;
-    pointer-events: none; border-radius: 50%;
-}
-
-@keyframes aurora {
-    from { transform: translateX(-50%) rotate(0deg); }
-    to   { transform: translateX(-50%) rotate(360deg); }
-}
-
-#hero::after {
-    content: ''; position: absolute;
-    bottom: 0; left: 8%; right: 8%; height: 1px;
-    background: linear-gradient(90deg, transparent, var(--b-2) 30%, var(--b-hi) 50%, var(--b-2) 70%, transparent);
-}
-
-.hero-corner { position: absolute; width: 36px; height: 36px; opacity: 0.35; }
-.hero-corner-tl { top: 18px; left: 18px; border-top: 1.5px solid var(--gold); border-left: 1.5px solid var(--gold); border-radius: 4px 0 0 0; }
-.hero-corner-tr { top: 18px; right: 18px; border-top: 1.5px solid var(--gold); border-right: 1.5px solid var(--gold); border-radius: 0 4px 0 0; }
-.hero-corner-bl { bottom: 18px; left: 18px; border-bottom: 1.5px solid var(--gold); border-left: 1.5px solid var(--gold); border-radius: 0 0 0 4px; }
-.hero-corner-br { bottom: 18px; right: 18px; border-bottom: 1.5px solid var(--gold); border-right: 1.5px solid var(--gold); border-radius: 0 0 4px 0; }
-
-.hero-eyebrow {
-    display: inline-flex; align-items: center; gap: 8px;
-    background: rgba(212,168,83,0.07); border: 1px solid rgba(212,168,83,0.2);
-    border-radius: 99px; padding: 5px 14px;
-    font-family: 'Geist Mono', monospace; font-size: 0.67rem;
-    color: var(--gold); letter-spacing: 2px; text-transform: uppercase;
-    margin-bottom: 20px; animation: fadeDown 0.6s ease both;
-}
-
-.hero-eyebrow::before {
-    content: ''; width: 5px; height: 5px; background: var(--gold);
-    border-radius: 50%; box-shadow: 0 0 6px var(--gold);
-    animation: pulse-dot 2s ease-in-out infinite;
-}
-
-@keyframes pulse-dot {
-    0%, 100% { opacity: 1; transform: scale(1); }
-    50% { opacity: 0.4; transform: scale(0.65); }
-}
-
-.hero-title {
-    font-family: 'Instrument Serif', Georgia, serif !important;
-    font-size: 3.4rem !important; font-weight: 400 !important;
-    color: var(--tx-1) !important; letter-spacing: -1.5px;
-    line-height: 1.05; margin-bottom: 16px !important;
-    animation: fadeDown 0.6s 0.1s ease both;
-}
-
-.hero-title em {
-    font-style: italic; color: var(--gold);
-    text-shadow: 0 0 40px rgba(212,168,83,0.3);
-}
-
-.hero-sub {
-    font-size: 1rem; color: var(--tx-2); font-weight: 300;
-    letter-spacing: 0.2px; max-width: 560px; margin: 0 auto 28px;
-    line-height: 1.6; animation: fadeDown 0.6s 0.2s ease both;
-}
-
-.pill-row {
-    display: flex; gap: 8px; justify-content: center;
-    flex-wrap: wrap; animation: fadeDown 0.6s 0.3s ease both;
-}
-
-.pill {
-    display: inline-flex; align-items: center; gap: 6px;
-    background: rgba(255,255,255,0.03); border: 1px solid var(--b-2);
-    border-radius: 99px; padding: 6px 14px;
-    font-size: 0.76rem; font-weight: 500;
-    color: var(--tx-2); letter-spacing: 0.2px;
-    transition: all 0.2s; cursor: default;
-}
-
-.pill:hover { border-color: var(--gold); color: var(--gold-lt); background: var(--gold-muted); transform: translateY(-1px); }
-.pill-dot { width: 5px; height: 5px; border-radius: 50%; flex-shrink: 0; }
-
-/* ═══════════ PANEL CARDS ═══════════ */
-.gr-group {
-    background: rgba(19,22,31,0.88) !important;
-    backdrop-filter: blur(20px) saturate(140%) !important;
-    -webkit-backdrop-filter: blur(20px) saturate(140%) !important;
-    border: 1px solid var(--b-1) !important;
-    border-radius: var(--r-lg) !important;
-    padding: 28px 32px !important;
-    box-shadow: var(--sh) !important;
-    transition: border-color 0.3s, box-shadow 0.3s !important;
-    position: relative; overflow: hidden;
-}
-
-.gr-group::before {
-    content: ''; position: absolute;
-    top: 0; left: 0; right: 0; height: 1px;
-    background: linear-gradient(90deg, transparent, rgba(255,255,255,0.05) 40%, rgba(255,255,255,0.05) 60%, transparent);
-    pointer-events: none;
-}
-
-.gr-group:hover { border-color: var(--b-2) !important; box-shadow: var(--sh-lg) !important; }
-
-/* ═══════════ STEP HEADERS ═══════════ */
-.step-header { display: flex; align-items: center; gap: 14px; margin-bottom: 20px; }
-
-.step-num {
-    width: 30px; height: 30px; background: var(--gold-muted);
-    border: 1px solid rgba(212,168,83,0.25); border-radius: 8px;
-    display: flex; align-items: center; justify-content: center;
-    font-family: 'Geist Mono', monospace; font-size: 0.72rem;
-    font-weight: 500; color: var(--gold); flex-shrink: 0;
-}
-
-.step-title { font-size: 0.95rem; font-weight: 600; color: var(--tx-1); letter-spacing: -0.2px; }
-.step-desc { font-size: 0.76rem; color: var(--tx-3); margin-left: auto; }
-.step-divider { height: 1px; background: linear-gradient(90deg, var(--b-1), transparent); margin-bottom: 20px; }
-
-/* ═══════════ FORM CONTROLS ═══════════ */
-label span, .block label span {
-    font-family: 'Geist', sans-serif !important; font-size: 0.73rem !important;
-    font-weight: 600 !important; color: var(--tx-3) !important;
-    letter-spacing: 0.8px !important; text-transform: uppercase !important;
-}
-
-textarea, input[type="text"], .gr-textbox textarea, .gr-textbox input {
-    background: var(--bg-3) !important; border: 1px solid var(--b-1) !important;
-    border-radius: var(--r-sm) !important; color: var(--tx-1) !important;
-    font-family: 'Geist Mono', monospace !important; font-size: 0.88rem !important;
-    padding: 14px 18px !important;
-    transition: border-color 0.2s, box-shadow 0.2s, background 0.2s !important;
-    resize: none !important; line-height: 1.5 !important;
-}
-
-textarea:focus, input[type="text"]:focus {
-    border-color: var(--gold) !important; box-shadow: var(--gold-ring) !important;
-    background: var(--bg-4) !important; outline: none !important;
-}
-
-textarea::placeholder, input::placeholder {
-    color: var(--tx-3) !important; font-style: italic;
-    font-family: 'Geist', sans-serif !important; font-size: 0.84rem !important;
-}
-
-select, .gr-dropdown select {
-    background: var(--bg-3) !important; border: 1px solid var(--b-1) !important;
-    border-radius: var(--r-sm) !important; color: var(--tx-1) !important;
-    font-family: 'Geist', sans-serif !important; font-size: 0.88rem !important;
-    padding: 14px 18px !important; transition: border-color 0.2s, box-shadow 0.2s !important;
-}
-
-select:focus { border-color: var(--gold) !important; box-shadow: var(--gold-ring) !important; outline: none !important; }
-
-/* ═══════════ BUTTONS ═══════════ */
-.gr-button {
-    font-family: 'Geist', sans-serif !important; font-size: 0.85rem !important;
-    font-weight: 600 !important; letter-spacing: 0.1px !important;
-    border-radius: var(--r-sm) !important; cursor: pointer !important;
-    transition: all 0.18s cubic-bezier(0.34, 1.56, 0.64, 1) !important;
-    padding: 13px 24px !important; position: relative; overflow: hidden;
-}
-
-.gr-button-primary::after {
-    content: ''; position: absolute;
-    top: 0; left: -100%; width: 100%; height: 100%;
-    background: linear-gradient(90deg, transparent, rgba(255,255,255,0.12), transparent);
-    transition: left 0.45s ease;
-}
-.gr-button-primary:hover::after { left: 100%; }
-
-.gr-button-primary {
-    background: linear-gradient(135deg, #d4a853 0%, #b8893a 60%, #9e7535 100%) !important;
-    color: #070810 !important; border: none !important;
-    box-shadow: 0 2px 16px rgba(212,168,83,0.3), inset 0 1px 0 rgba(255,255,255,0.15) !important;
-}
-
-.gr-button-primary:hover {
-    transform: translateY(-2px) scale(1.01) !important;
-    box-shadow: 0 8px 28px rgba(212,168,83,0.42), inset 0 1px 0 rgba(255,255,255,0.2) !important;
-    filter: brightness(1.06) !important;
-}
-
-.gr-button-primary:active { transform: translateY(0) scale(0.99) !important; }
-
-.gr-button-secondary {
-    background: var(--bg-4) !important; color: var(--tx-2) !important;
-    border: 1px solid var(--b-2) !important;
-}
-
-.gr-button-secondary:hover {
-    border-color: var(--gold) !important; color: var(--gold-lt) !important;
-    background: var(--gold-muted) !important; transform: translateY(-1px) !important;
-    box-shadow: var(--sh-gold) !important;
-}
-
-/* ═══════════ STATUS BOX ═══════════ */
-#status-box .prose p, #status-box p {
-    font-family: 'Geist Mono', monospace !important; font-size: 0.8rem !important;
-    color: var(--tx-2) !important; background: var(--bg-3) !important;
-    border: 1px solid var(--b-1) !important; border-left: 3px solid var(--gold) !important;
-    border-radius: 0 var(--r-xs) var(--r-xs) 0 !important;
-    padding: 11px 16px !important; margin: 0 !important; line-height: 1.55 !important;
-}
-
-/* ═══════════ SUMMARY OUTPUT ═══════════ */
-#summary-out .prose strong, #summary-out strong { color: var(--gold-lt) !important; font-weight: 600 !important; }
-#summary-out .prose p { margin-bottom: 10px !important; }
-#summary-out .prose ul { padding-left: 20px !important; }
-#summary-out .prose li { margin-bottom: 5px !important; color: var(--tx-1) !important; }
-#summary-out .prose em { color: var(--tx-2) !important; }
-
-/* ═══════════ CHATBOT ═══════════ */
-.gr-chatbot, [data-testid="chatbot"] {
-    background: var(--bg-1) !important; border: 1px solid var(--b-1) !important;
-    border-radius: var(--r) !important; padding: 12px !important;
-}
-
-.gr-chatbot .message {
-    font-family: 'Geist', sans-serif !important; font-size: 0.9rem !important;
-    line-height: 1.7 !important; padding: 13px 18px !important;
-    border-radius: 12px !important; max-width: 82% !important;
-    animation: msgIn 0.25s cubic-bezier(0.34,1.56,0.64,1) both;
-}
-
-@keyframes msgIn {
-    from { opacity: 0; transform: translateY(8px) scale(0.97); }
-    to   { opacity: 1; transform: translateY(0) scale(1); }
-}
-
-.gr-chatbot .message.user {
-    background: linear-gradient(135deg, var(--bg-4), var(--bg-5)) !important;
-    border: 1px solid var(--b-2) !important; color: var(--tx-1) !important; margin-left: auto !important;
-}
-
-.gr-chatbot .message.bot {
-    background: var(--bg-2) !important; border: 1px solid var(--b-1) !important; color: var(--tx-1) !important;
-}
-
-/* ═══════════ AUDIO ═══════════ */
-.gr-audio, [data-testid="audio"] {
-    background: var(--bg-3) !important; border: 1px solid var(--b-1) !important;
-    border-radius: var(--r-sm) !important; padding: 8px !important;
-}
-
-/* ═══════════ VOICE DESC ═══════════ */
-.voice-desc {
-    font-size: 0.84rem; color: var(--tx-2); margin: 0 0 16px; line-height: 1.6;
-    padding: 10px 14px; background: var(--bg-3); border-radius: var(--r-xs);
-    border-left: 2px solid var(--b-2);
-}
-
-/* ═══════════ FOOTER ═══════════ */
-#footer-bar {
-    display: flex; align-items: center; justify-content: space-between;
-    padding: 14px 20px; background: rgba(15,17,25,0.6);
-    border: 1px solid var(--b-0); border-radius: var(--r); margin-top: 4px;
-    flex-wrap: wrap; gap: 10px;
-}
-
-.footer-copy { font-family: 'Geist Mono', monospace; font-size: 0.68rem; color: var(--tx-3); }
-.footer-stack { display: flex; gap: 6px; flex-wrap: wrap; }
-.footer-tag {
-    font-family: 'Geist Mono', monospace; font-size: 0.64rem; color: var(--tx-3);
-    background: var(--bg-2); border: 1px solid var(--b-0); border-radius: 4px; padding: 2px 7px;
-}
-
-/* ═══════════ SCROLLBAR ═══════════ */
-::-webkit-scrollbar { width: 4px; height: 4px; }
-::-webkit-scrollbar-track { background: transparent; }
-::-webkit-scrollbar-thumb { background: var(--b-2); border-radius: 99px; }
-::-webkit-scrollbar-thumb:hover { background: var(--gold-dk); }
-
-/* ═══════════ ANIMATIONS ═══════════ */
-@keyframes fadeDown { from { opacity: 0; transform: translateY(-10px); } to { opacity: 1; transform: translateY(0); } }
-@keyframes fadeUp   { from { opacity: 0; transform: translateY(16px);  } to { opacity: 1; transform: translateY(0); } }
-@keyframes fadeIn   { from { opacity: 0; } to { opacity: 1; } }
-
-.gradio-container .gr-group { animation: fadeUp 0.5s cubic-bezier(0.22, 1, 0.36, 1) both; }
-
-::selection { background: rgba(212,168,83,0.25); color: var(--tx-1); }
-"""
+with gr.Blocks() as demo:
 
-# =========================
-# BUILD UI
-# =========================
+    gr.Markdown("# 📚 ArXiv Research Assistant", elem_id="title")
 
-with gr.Blocks(title="Research Intelligence Platform") as demo:
-
-    # ─── HERO HEADER ──────────────────────────────────────────────────
-    gr.HTML("""
-    <div id="hero">
-        <div class="hero-corner hero-corner-tl"></div>
-        <div class="hero-corner hero-corner-tr"></div>
-        <div class="hero-corner hero-corner-bl"></div>
-        <div class="hero-corner hero-corner-br"></div>
-
-        <div class="hero-eyebrow">Research Intelligence Platform</div>
-
-        <div class="hero-title">Your <em>AI Research</em> Partner</div>
-
-        <div class="hero-sub">
-            Paste any paper DOI — instantly fetch, index, summarize, and interrogate
-            research literature with LLaMA 3.3 and semantic search.
-        </div>
-
-        <div class="pill-row">
-            <span class="pill"><span class="pill-dot" style="background:#3ecf8e"></span>RAG Architecture</span>
-            <span class="pill"><span class="pill-dot" style="background:#d4a853"></span>LLaMA 3.3 · 70B</span>
-            <span class="pill"><span class="pill-dot" style="background:#4d94ff"></span>Whisper ASR</span>
-            <span class="pill"><span class="pill-dot" style="background:#f0a443"></span>FAISS Semantic Index</span>
-            <span class="pill"><span class="pill-dot" style="background:#c084fc"></span>DOI Resolver</span>
-        </div>
-    </div>
-    """)
-
-    # ─── STEP 01 — LOAD PAPER ─────────────────────────────────────────
-    with gr.Group():
-        gr.HTML("""
-        <div class="step-header">
-            <div class="step-num">01</div>
-            <div><div class="step-title">Load Paper</div></div>
-            <div class="step-desc">Paste DOI &rarr; fetch &rarr; index</div>
-        </div>
-        <div class="step-divider"></div>
-        """)
-        with gr.Row(equal_height=True):
-            arxiv_input = gr.Textbox(
-                label="Paper DOI",
-                placeholder="10.1038/s41586-021-03819-2  ·  https://doi.org/10.48550/arXiv.1706.03762",
-                scale=5,
-            )
-            load_btn = gr.Button("Load Paper →", variant="primary", scale=1, min_width=160)
-        status = gr.Markdown(
-            value="*Paste a DOI above and click **Load Paper** — the full text will be fetched and indexed automatically.*",
-            elem_id="status-box",
-        )
+    with gr.Row():
+        arxiv_input = gr.Textbox(label="Enter arXiv ID", scale=4)
+        load_btn = gr.Button("Load Paper", variant="primary", scale=1)
 
-    # ─── STEP 02 — SECTION SUMMARY ────────────────────────────────────
-    with gr.Group():
-        gr.HTML("""
-        <div class="step-header">
-            <div class="step-num">02</div>
-            <div><div class="step-title">Section Summary</div></div>
-            <div class="step-desc">AI-powered breakdown</div>
-        </div>
-        <div class="step-divider"></div>
-        """)
-        with gr.Row(equal_height=True):
-            section_dropdown = gr.Dropdown(
-                label="Select Section", choices=[], scale=5, interactive=True,
-            )
-            summarize_btn = gr.Button("✦ Summarize", variant="secondary", scale=1, min_width=150)
-        summary_output = gr.Markdown(
-            value="*Select a section from the dropdown, then click **Summarize** for a structured AI breakdown.*",
-            elem_id="summary-out",
-        )
+    status = gr.Markdown()
+
+    with gr.Row():
+        section_dropdown = gr.Dropdown(label="Sections", scale=3)
+        summarize_btn = gr.Button("Generate Summary", variant="secondary", scale=1)
+
+    summary_output = gr.Markdown()
 
-    # ─── STEP 03 — CHAT ───────────────────────────────────────────────
-    with gr.Group():
-        gr.HTML("""
-        <div class="step-header">
-            <div class="step-num">03</div>
-            <div><div class="step-title">Chat with Paper</div></div>
-            <div class="step-desc">Context-aware Q&A</div>
-        </div>
-        <div class="step-divider"></div>
-        """)
-        chatbot = gr.Chatbot(value=[], height=440, show_label=False)
-        with gr.Row(equal_height=True):
-            msg = gr.Textbox(
-                label="Ask a Question",
-                placeholder="What is the main contribution?  ·  How does the method work?  ·  What datasets were used?",
-                scale=5, lines=1, max_lines=4,
-            )
-            send_btn = gr.Button("Send →", variant="primary", scale=1, min_width=120)
-
-    # ─── STEP 04 — VOICE ──────────────────────────────────────────────
-    with gr.Group():
-        gr.HTML("""
-        <div class="step-header">
-            <div class="step-num">04</div>
-            <div><div class="step-title">Voice Query</div></div>
-            <div class="step-desc">Speak &rarr; transcribe &rarr; search</div>
-        </div>
-        <div class="step-divider"></div>
-        <p class="voice-desc">
-            Record or upload an audio question — Whisper ASR transcribes it and runs semantic search automatically.
-        </p>
-        """)
-        with gr.Row(equal_height=True):
-            audio = gr.Audio(type="filepath", label="Record or Upload Audio", scale=5)
-            voice_btn = gr.Button("🎙 Transcribe & Ask", variant="secondary", scale=1, min_width=170)
-
-    # ─── FOOTER ───────────────────────────────────────────────────────
-    gr.HTML("""
-    <div id="footer-bar">
-        <span class="footer-copy">Research Intelligence Platform &middot; 2026</span>
-        <div class="footer-stack">
-            <span class="footer-tag">Gradio</span>
-            <span class="footer-tag">Groq LLaMA 3.3</span>
-            <span class="footer-tag">FAISS</span>
-            <span class="footer-tag">Sentence Transformers</span>
-            <span class="footer-tag">faster-whisper</span>
-            <span class="footer-tag">PyMuPDF</span>
-            <span class="footer-tag">Unpaywall</span>
-        </div>
-    </div>
-    """)
-
-    # ─── BINDINGS ─────────────────────────────────────────────────────
+    gr.Markdown("## 💬 Chat with Paper")
+    chatbot = gr.Chatbot(height=400)
+
+    with gr.Row():
+        msg = gr.Textbox(label="Ask a question", scale=4)
+        send_btn = gr.Button("Send", variant="primary", scale=1)
+
+    gr.Markdown("## 🎙 Voice Query")
+
+    with gr.Row():
+        audio = gr.Audio(type="filepath", scale=4)
+        voice_btn = gr.Button("Ask via Voice", scale=1)
+
+    # Actions
     load_btn.click(load_paper, inputs=arxiv_input, outputs=[section_dropdown, status])
     summarize_btn.click(summarize_section, inputs=section_dropdown, outputs=summary_output)
     send_btn.click(rag_chat, inputs=[msg, chatbot], outputs=[chatbot, msg])
-    msg.submit(rag_chat, inputs=[msg, chatbot], outputs=[chatbot, msg])
     voice_btn.click(voice_chat, inputs=[audio, chatbot], outputs=[chatbot, msg])
 
-demo.launch(css=CSS)
+
+demo.launch(
+    theme=gr.themes.Soft(),
+    css="""
+    #title {text-align:center}
+    """
+)