TrendingAIBot / app.py
Sumayyea's picture
Update app.py
f1176e0 verified
Raw History Blame Contribute Delete
13.5 kB
"""
AI Trends Assistant - Fetches trending AI papers (arXiv) and YouTube videos
"""
import os
import streamlit as st
import arxiv
import requests
from dotenv import load_dotenv
from typing import List, Dict
load_dotenv()
# ---------------------
# API Keys
# ---------------------
GEMINI_API_KEY = os.environ.get("GEMINI_API_KEY")
GEMINI_API_URL = os.environ.get("GEMINI_API_URL", "").strip()
YOUTUBE_API_KEY = os.environ.get("YOUTUBE_API_KEY")
# ---------------------
# Super Prompt
# ---------------------
SYSTEM_PROMPT = """
Role: You are an expert AI research assistant who specializes in discovering and explaining trending topics, papers, and videos about Artificial Intelligence.
Goal: Help users explore and understand what's currently trending in AI.
"""
# ---------------------
# Gemini API wrapper
# ---------------------
def call_gemini(system_prompt: str, user_message: str, model: str = "gemini-2.0-flash") -> str:
if not GEMINI_API_KEY:
return "[Gemini unavailable] No GEMINI_API_KEY found."
if not GEMINI_API_URL:
return "[Gemini endpoint not configured] Set GEMINI_API_URL."
try:
headers = {
"Authorization": f"Bearer {GEMINI_API_KEY}",
"Content-Type": "application/json",
}
payload = {"system": system_prompt, "input": user_message, "model": model}
resp = requests.post(GEMINI_API_URL, headers=headers, json=payload, timeout=30)
resp.raise_for_status()
data = resp.json()
if isinstance(data, dict):
if "output" in data:
return data["output"]
if "choices" in data and data["choices"]:
ch0 = data["choices"][0]
if "text" in ch0:
return ch0["text"]
if "message" in ch0 and "content" in ch0["message"]:
return ch0["message"]["content"]
return str(data)
except Exception as e:
return f"[Gemini request failed] {e}"
# ---------------------
# arXiv fetcher (IMPROVED)
# ---------------------
def fetch_arxiv(query: str, max_results: int = 5, sort_by: str = "Relevance") -> List[Dict]:
"""
Fetch AI papers from arXiv with improved search relevance.
Improvements:
- Searches in title (ti:) and abstract (abs:) for better matching
- Supports sorting by Relevance or Recent
- Uses quoted phrases for multi-word queries
- Filters to CS/AI-related categories
"""
# Clean and prepare query
query = query.strip()
# Build search query - search in title and abstract
# For multi-word queries, search as phrase AND individual terms
words = query.split()
if len(words) > 1:
# Multi-word: search exact phrase in title OR abstract, plus individual terms
phrase_query = f'"{query}"'
title_search = f'ti:{phrase_query}'
abstract_search = f'abs:{phrase_query}'
# Also search individual important words (for broader matching)
word_searches = " AND ".join([f'all:{w}' for w in words if len(w) > 2])
# Combine: (exact phrase in title OR abstract) OR (all words present)
search_query = f'({title_search} OR {abstract_search}) OR ({word_searches})'
else:
# Single word: search in title and abstract
search_query = f'ti:{query} OR abs:{query}'
# Add AI-related category filter (but use OR to not be too restrictive)
ai_categories = "(cat:cs.AI OR cat:cs.LG OR cat:cs.CL OR cat:cs.CV OR cat:cs.NE OR cat:cs.MA OR cat:stat.ML)"
full_query = f'({search_query}) AND {ai_categories}'
# Determine sort criteria
if sort_by == "Relevance":
sort_criterion = arxiv.SortCriterion.Relevance
else:
sort_criterion = arxiv.SortCriterion.SubmittedDate
search = arxiv.Search(
query=full_query,
max_results=max_results * 2, # Fetch extra to filter duplicates
sort_by=sort_criterion,
sort_order=arxiv.SortOrder.Descending
)
results = []
seen_titles = set()
try:
for r in search.results():
# Skip duplicates
title_lower = r.title.lower().strip()
if title_lower in seen_titles:
continue
seen_titles.add(title_lower)
# Calculate a simple relevance indicator
title_match = query.lower() in r.title.lower()
abstract_match = query.lower() in r.summary.lower()
results.append({
"title": r.title,
"url": r.entry_id,
"pdf_url": r.pdf_url,
"summary": r.summary[:400] + "..." if len(r.summary) > 400 else r.summary,
"published": r.published.strftime("%Y-%m-%d") if r.published else "N/A",
"updated": r.updated.strftime("%Y-%m-%d") if r.updated else "N/A",
"authors": ", ".join([a.name for a in r.authors[:4]]) + ("..." if len(r.authors) > 4 else ""),
"categories": ", ".join(r.categories[:3]),
"title_match": title_match,
"abstract_match": abstract_match
})
if len(results) >= max_results:
break
except Exception as e:
st.error(f"arXiv API error: {e}")
# Sort results: prioritize title matches, then abstract matches
results.sort(key=lambda x: (x["title_match"], x["abstract_match"]), reverse=True)
return results
def fetch_arxiv_simple(query: str, max_results: int = 5) -> List[Dict]:
"""
Fallback simple search if advanced search returns no results.
"""
search = arxiv.Search(
query=query,
max_results=max_results,
sort_by=arxiv.SortCriterion.Relevance,
sort_order=arxiv.SortOrder.Descending
)
results = []
try:
for r in search.results():
results.append({
"title": r.title,
"url": r.entry_id,
"pdf_url": r.pdf_url,
"summary": r.summary[:400] + "..." if len(r.summary) > 400 else r.summary,
"published": r.published.strftime("%Y-%m-%d") if r.published else "N/A",
"updated": r.updated.strftime("%Y-%m-%d") if r.updated else "N/A",
"authors": ", ".join([a.name for a in r.authors[:4]]) + ("..." if len(r.authors) > 4 else ""),
"categories": ", ".join(r.categories[:3]),
"title_match": False,
"abstract_match": False
})
except Exception as e:
st.error(f"arXiv API error: {e}")
return results
# ---------------------
# YouTube fetcher (Official API)
# ---------------------
def fetch_youtube(query: str, max_results: int = 5) -> List[Dict]:
"""Fetch YouTube videos using official YouTube Data API v3."""
if not YOUTUBE_API_KEY:
return []
try:
url = "https://www.googleapis.com/youtube/v3/search"
params = {
"part": "snippet",
"q": f"{query} AI tutorial",
"type": "video",
"maxResults": max_results,
"order": "relevance",
"key": YOUTUBE_API_KEY
}
response = requests.get(url, params=params, timeout=10)
response.raise_for_status()
data = response.json()
videos = []
for item in data.get("items", []):
video_id = item["id"]["videoId"]
snippet = item["snippet"]
videos.append({
"title": snippet.get("title", "Unknown"),
"url": f"https://www.youtube.com/watch?v={video_id}",
"channel": snippet.get("channelTitle", "Unknown"),
"description": snippet.get("description", "")[:150] + "...",
"thumbnail": snippet.get("thumbnails", {}).get("medium", {}).get("url", "")
})
return videos
except Exception as e:
st.error(f"YouTube API error: {e}")
return []
# ---------------------
# Streamlit UI
# ---------------------
st.set_page_config(page_title="AI Trends Assistant", layout="centered")
st.title("AI Trends Assistant")
st.markdown("Fetch trending AI papers from arXiv and related YouTube videos.")
with st.sidebar:
st.header("βš™οΈ Settings")
mode = st.radio("Mode", ["Fetch Links", "Chat (Gemini)"])
max_results = st.slider("Max results", min_value=1, max_value=20, value=5)
st.markdown("---")
st.subheader("arXiv Settings")
arxiv_sort = st.radio("Sort papers by:", ["Relevance", "Recent"], index=0)
st.markdown("---")
st.markdown("**API Status:**")
st.markdown(f"YouTube API: {'βœ… Connected' if YOUTUBE_API_KEY else '❌ Not configured'}")
st.markdown(f"Gemini API: {'βœ… Connected' if GEMINI_API_KEY else '❌ Not configured'}")
if mode == "Fetch Links":
st.subheader("πŸ” Search Trending AI Topics")
trending_topics = [
"Custom query...",
"generative AI",
"large language models",
"multimodal learning",
"diffusion models",
"reinforcement learning from human feedback",
"vision transformer",
"retrieval augmented generation",
"graph neural networks",
"federated learning",
"AI agents",
"neural radiance fields",
"text to image generation",
"chain of thought reasoning",
"mixture of experts"
]
selected_topic = st.selectbox("Select a trending topic:", trending_topics)
if selected_topic == "Custom query...":
user_query = st.text_input("Enter your search query:", value="", placeholder="e.g., generative AI, transformer models, etc.")
else:
user_query = selected_topic
col1, col2 = st.columns(2)
fetch_arxiv_btn = col1.checkbox("Fetch arXiv Papers", value=True)
fetch_youtube_btn = col2.checkbox("Fetch YouTube Videos", value=True)
if st.button("πŸš€ Fetch Results"):
if not user_query.strip():
st.warning("Please enter or select a query.")
else:
# arXiv Results
if fetch_arxiv_btn:
st.write("### πŸ“„ arXiv Papers")
st.caption(f"Searching for: **{user_query}** | Sorted by: **{arxiv_sort}**")
with st.spinner("Fetching papers..."):
papers = fetch_arxiv(user_query, max_results=max_results, sort_by=arxiv_sort)
# Fallback to simple search if no results
if not papers:
st.info("Trying broader search...")
papers = fetch_arxiv_simple(user_query, max_results=max_results)
if not papers:
st.warning("No arXiv results found. Try different keywords.")
else:
for i, p in enumerate(papers, 1):
# Show relevance badge
badge = ""
if p.get("title_match"):
badge = "🎯 "
elif p.get("abstract_match"):
badge = "βœ“ "
with st.expander(f"{badge}{i}. {p['title']}", expanded=(i <= 3)):
st.markdown(f"**Authors:** {p['authors']}")
st.markdown(f"**Published:** {p['published']} | **Updated:** {p['updated']}")
st.markdown(f"**Categories:** `{p['categories']}`")
st.markdown(f"**Abstract:** {p['summary']}")
col1, col2 = st.columns(2)
col1.markdown(f"πŸ“„ [View Paper]({p['url']})")
col2.markdown(f"πŸ“₯ [Download PDF]({p['pdf_url']})")
# YouTube Results
if fetch_youtube_btn:
st.write("### 🎬 YouTube Videos")
if not YOUTUBE_API_KEY:
st.warning("⚠️ YouTube API key not configured. Add `YOUTUBE_API_KEY` to your Hugging Face Secrets.")
else:
with st.spinner("Fetching videos..."):
videos = fetch_youtube(user_query, max_results=max_results)
if not videos:
st.info("No YouTube results found.")
else:
for v in videos:
col1, col2 = st.columns([1, 3])
with col1:
if v.get("thumbnail"):
st.image(v["thumbnail"], width=120)
with col2:
st.markdown(f"**[{v['title']}]({v['url']})**")
st.caption(f"πŸ“Ί {v['channel']}")
else:
st.subheader("πŸ’¬ Chat with AI Trends Assistant")
user_msg = st.text_area("Your message:", value="What's trending in multimodal AI research?")
if st.button("Send"):
with st.spinner("Calling Gemini..."):
reply = call_gemini(SYSTEM_PROMPT, user_msg)
st.markdown("**Assistant:**")
st.write(reply)
st.markdown("---")
st.caption("Data sources: arXiv API, YouTube Data API v3")