NewsFindr – LLM Powered Personalized News Agent

AI-powered system that retrieves personalized and credible news articles using an LLM-based agent with query expansion, web search, credibility filtering, and automated summarization.

Problem

Searching for relevant and reliable news is challenging because:

• Users receive generic results not tailored to their interests
• Search engines often return duplicate or low-credibility sources
• Important information is spread across multiple articles
• Users must manually read several articles to understand the overall story
• An intelligent system is needed to personalize, filter, and summarize news automatically.

Approach

The system implements an LLM-powered AI agent pipeline:

• Personalization: Fetches user interests using email from a database
• Query Expansion: Uses an LLM to generate multiple related search queries
• Web Search: Retrieves news articles using DuckDuckGo
• Credibility Filtering: Filters irrelevant or low-quality sources
• Summarization: Generates a concise summary of key news insights

Pipeline

User Email + Query
        ↓
Fetch User Interests (Database)
        ↓
AI Agent (Tool Calling + Reasoning)
        ↓
Query Expansion
        ↓
Web Search (DuckDuckGo)
        ↓
Credibility Filtering
        ↓
News Summarization
        ↓
Personalized News Insights

Results

• Personalized news retrieval based on user interests
• Improved search coverage using LLM query expansion
• Reduced noise through filtering
• Generated concise summaries from multiple sources
• Demonstrates a practical AI agent-based retrieval system

Tech Stack

• Python
• Groq API (LLM)
• LangChain (Agents + Tools)
• DuckDuckGo Search
• SQLite

Project Structure

newsfindr/
│
├── newsfindr.py
├── customer.db
├── requirements.txt
├── README.md

Setup

Clone the repository

git clone https://github.com/your-username/newsfindr.git
cd newsfindr

Install dependencies

pip install -r requirements.txt

Set your API key

Linux / Mac:

export GROQ_API_KEY="your_api_key_here"

Windows:

set GROQ_API_KEY=your_api_key_here

Run the project

python newsfindr.py

Example

• Input: User email and query

• Output:

• List of relevant news articles, AI-generated summary

{
  "summary": "...",
  "urls": [...]
}

Limitations

• Summaries are generated from search results (metadata/snippets), not full article content
• Credibility filtering is heuristic and may not always be accurate
• Depends on DuckDuckGo search results

Future Improvements

• Add full article extraction (e.g., newspaper3k)
• Build a streamlit dashboard
• Implement source credibility scoring