NewsWise - Intelligent News Q & A Agent

An intelligent news Q&A agent built with LangChain and LangGraph. Using RAG (Retrieval Augmented Generation) to answer questions from BBC New articles (Source: Kaggle) with conversation memory.

Heads up! This repo is purely for learning purposes, just me exploring and experimenting. Nothing fancy here! 😄

Features

RAG Pipeline: Retrieves relevant BBC news articles using FAISS vector search
LLM-powered answer: Using Claude (Anthropic) model to answer question from retrieved context (from RAG's retrieval)
Conversation Memory: Remember previous question and answer across turns
LangGraph Flow: Built with StateGraph production ready workflow
MemorySaver: Persist graph state between interactions
HuggingFaceEmbeddings: for semantic search

Architecture

  User Question
       ↓
  [LangGraph StateGraph]
       ↓
  [ask] → get user input
       ↓
  [route] → /exit or continue
       ↓
  [answer] → RAG Chain
       ├── FAISS Retriever (top 3 articles)
       ├── ChatPromptTemplate + History
       └── Claude LLM → StrOutputParser
       ↓
  Loop back to [ask]

Dataset

Source: BBC News Archive - Kaggle
Size: 2,225 articles
Categories: Business, Sport, Politics, Tech, Entertainment
Columns: category, filename, title, content

Tech Stack

Tool	Purpose
LangChain	LLM chaining, prompt templates, RAG
LangGraph	Agent flow, state management
Anthropic Claude	LLM for answer generation
FAISS	Vector store for semantic search
HuggingFace	Sentence embeddings
Pandas	Dataset loading and processing

Setup

Clone the repository

git clone https://github.com/yourusername/newswise.git
cd newswise

Install dependencies

pip install langchain langchain-anthropic langchain-community langchain-huggingface faiss-cpu langgraph sentence-transformers pandas

Set up API keys Add to your environment or Google Colab secrets:

ANTHROPIC_API_KEY=your_anthropic_api_key
KAGGLE_USERNAME=your_kaggle_username
KAGGLE_KEY=your_kaggle_api_key

Download dataset

kaggle datasets download -d hgultekin/bbcnewsarchive
unzip bbcnewsarchive.zip

Run the notebook Open newswise.ipynb in Google Colab or Jupyter and run all cells.

Usage

Enter a Question: What happened in football?
Agent: Based on the provided context...

Enter a Question: Who was mentioned in the last answer?
Agent: Based on the previous answer...

Enter a Question: /exit

Type any news-related question
Type /exit to end the session

Future Improvements

Add category filter tool
Swap MemorySaver to PostgreSQL for production
Add Gradio UI for web interface
Add source citation in answers
Deploy to Hugging Face Spaces

License

MIT License: free to use, modify and share.

Acknowledgements

BBC News Archive Dataset by Habib Gültekin on Kaggle
LangChain and LangGraph teams
Anthropic for Claude

RPG-coder/kaggle-agentic-ai

README