Update app.py

app.py

@@ -1,36 +1,168 @@
 import gradio as gr
+import json
+import requests
+import urllib.request
+import urllib.parse
+import xml.etree.ElementTree as ET
+import re
 from huggingface_hub import hf_hub_download
 from llama_cpp import Llama
-import spaces
+from tavily import TavilyClient
+from duckduckgo_search import DDGS
 
-# 1. Download NexAI GGUF model directly from your Hugging Face Repo
-print("📥 Downloading/Loading NexAI-v1 GGUF...")
+# ==============================================================================
+# 1. API KEYS CONFIGURATION
+# ==============================================================================
+TAVILY_API_KEY = "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"
+SERPER_API_KEY = "934f29a2bff6c477732baebb7e208e57d43303e7"
+
+# Initialize Tavily
+try:
+    tavily_client = TavilyClient(api_key=TAVILY_API_KEY)
+except Exception:
+    tavily_client = None
+
+# ==============================================================================
+# 2. 4-TIER MULTI-SEARCH FALLBACK ENGINE
+# ==============================================================================
+def search_tier1_tavily(query: str):
+    """Tier 1: Tavily AI-Powered Fact Extraction"""
+    if not tavily_client:
+        raise Exception("Tavily client not ready")
+    response = tavily_client.search(query=query, search_depth="basic", max_results=3)
+    snippets = []
+    for r in response.get("results", []):
+        snippets.append(f"• {r.get('title')}: {r.get('content')}")
+    if snippets:
+        return "[Source: Tavily AI Search]\n" + "\n".join(snippets)
+    raise Exception("Tavily returned 0 results")
+
+def search_tier2_serper_google(query: str):
+    """Tier 2: Serper.dev Official Google Search"""
+    url = "https://google.serper.dev/search"
+    headers = {
+        "X-API-KEY": SERPER_API_KEY,
+        "Content-Type": "application/json"
+    }
+    payload = json.dumps({"q": query, "gl": "in", "hl": "hi"})
+    res = requests.post(url, headers=headers, data=payload, timeout=6)
+    if res.status_code == 200:
+        data = res.json()
+        snippets = []
+        if "knowledgeGraph" in data:
+            kg = data["knowledgeGraph"]
+            snippets.append(f"• Fact: {kg.get('title')} - {kg.get('description', '')}")
+        for item in data.get("organic", [])[:3]:
+            snippets.append(f"• {item.get('title')}: {item.get('snippet')}")
+        if snippets:
+            return "[Source: Google Serper]\n" + "\n".join(snippets)
+    raise Exception(f"Serper error status {res.status_code}")
+
+def search_tier3_duckduckgo(query: str):
+    """Tier 3: DuckDuckGo Fallback"""
+    snippets = []
+    with DDGS() as ddgs:
+        for r in ddgs.text(query, max_results=3):
+            snippets.append(f"• {r.get('title')}: {r.get('body')}")
+    if snippets:
+        return "[Source: DuckDuckGo]\n" + "\n".join(snippets)
+    raise Exception("DuckDuckGo failed")
+
+def search_tier4_google_news_rss(query: str):
+    """Tier 4: Unlimited Free Google News RSS Feed (No API Key Required)"""
+    encoded_query = urllib.parse.quote(query)
+    url = f"https://news.google.com/rss/search?q={encoded_query}&hl=en-IN&gl=IN&ceid=IN:en"
+    req = urllib.request.Request(url, headers={"User-Agent": "Mozilla/5.0"})
+    with urllib.request.urlopen(req, timeout=6) as response:
+        xml_data = response.read()
+        root = ET.fromstring(xml_data)
+        items = []
+        for item in root.findall("./channel/item")[:4]:
+            title = item.find("title").text
+            pub_date = item.find("pubDate").text if item.find("pubDate") is not None else ""
+            items.append(f"• [{pub_date}] {title}")
+        if items:
+            return "[Source: Google News Live RSS]\n" + "\n".join(items)
+    raise Exception("Google News RSS failed")
+
+def multi_tier_search(query: str) -> str:
+    """Cascade through Tier 1 -> Tier 2 -> Tier 3 -> Tier 4"""
+    # 1. Try Tavily
+    try:
+        return search_tier1_tavily(query)
+    except Exception as e1:
+        print(f"Tier 1 (Tavily) failed: {e1}")
+
+    # 2. Try Serper (Google)
+    try:
+        return search_tier2_serper_google(query)
+    except Exception as e2:
+        print(f"Tier 2 (Serper) failed: {e2}")
+
+    # 3. Try DuckDuckGo
+    try:
+        return search_tier3_duckduckgo(query)
+    except Exception as e3:
+        print(f"Tier 3 (DuckDuckGo) failed: {e3}")
+
+    # 4. Try Free Google News RSS
+    try:
+        return search_tier4_google_news_rss(query)
+    except Exception as e4:
+        print(f"Tier 4 (Google News RSS) failed: {e4}")
+
+    return "Real-time search temporarily unavailable."
+
+# ==============================================================================
+# 3. LOAD NEXAI GGUF MODEL
+# ==============================================================================
+print("📥 Loading NexAI-v1 GGUF Model...")
 model_path = hf_hub_download(
     repo_id="Anoopsingh53/nexai-v1",
     filename="nexai-v1-Q4_K_M.gguf"
 )
 
-# 2. Initialize Llama model
 llm = Llama(
     model_path=model_path,
-    n_ctx=2048,
+    n_ctx=3072,
     n_threads=4
 )
 
-SYSTEM_PROMPT = """You are NexAI, an intelligent, high-speed AI assistant fine-tuned for reasoning, coding, and real-time conversation. Answer clearly and politely in Hindi/English."""
+def is_search_needed(query: str) -> bool:
+    """Auto-detect if query asks for live/real-time information"""
+    keywords = [
+        "today", "latest", "aaj", "current", "news", "price", "weather", 
+        "match", "score", "update", "2026", "2025", "who won", "koun",
+        "stock", "status", "recent", "kab", "kya chal raha", "taaza"
+    ]
+    q = query.lower()
+    return any(kw in q for kw in keywords)
 
-# 3. Chat generation function
+# ==============================================================================
+# 4. CHAT GENERATION WITH REAL-TIME CONTEXT
+# ==============================================================================
 def predict(message, history):
-    messages = [{"role": "system", "content": SYSTEM_PROMPT}]
+    live_context = ""
+    
+    if is_search_needed(message):
+        yield "🔍 *Searching live web with multi-tier fallback (Tavily / Serper / Google)...*\n"
+        search_data = multi_tier_search(message)
+        live_context = f"\n\n[REAL-TIME LIVE DATA]:\n{search_data}\n\nBased on the above real-time data, provide a direct, accurate answer."
+
+    system_prompt = f"""You are NexAI, an intelligent, real-time AI assistant. 
+Answer politely and clearly in Hindi or English using provided real-time data.{live_context}"""
+
+    messages = [{"role": "system", "content": system_prompt}]
     for user_msg, bot_msg in history:
+        clean_bot = re.sub(r"🔍 \*.*?\*\n", "", bot_msg)
         messages.append({"role": "user", "content": user_msg})
-        messages.append({"role": "assistant", "content": bot_msg})
+        messages.append({"role": "assistant", "content": clean_bot})
     messages.append({"role": "user", "content": message})
 
     response = llm.create_chat_completion(
         messages=messages,
         max_tokens=512,
-        temperature=0.7,
+        temperature=0.6,
         stream=True
     )
 
@@ -41,12 +173,19 @@ def predict(message, history):
             partial_text += delta["content"]
             yield partial_text
 
-# 4. Gradio Chat Interface
+# ==============================================================================
+# 5. GRADIO UI
+# ==============================================================================
 demo = gr.ChatInterface(
     fn=predict,
-    title="🚀 NexAI-v1 Live Agent & API",
-    description="Live inference powered by NexAI-v1 GGUF.",
-    examples=["Who are you?", "Explain Quantum Computing in Hinglish", "Write a Python script for web search"],
+    title="🚀 NexAI-v1 — Real-Time AI Agent",
+    description="Powered by NexAI 3B GGUF + Multi-Tier Search (Tavily ➔ Serper Google ➔ DuckDuckGo ➔ Google News RSS).",
+    examples=[
+        "Tell me today's latest tech & AI news",
+        "What is the current status of ISRO Gaganyaan mission?",
+        "Who is the current Prime Minister of India?",
+        "Explain Neural Networks in simple Hinglish"
+    ],
 )
 
 if __name__ == "__main__":