NEWSMIND AI: A Retrieval-Augmented Generation System for Intelligent News Analysis

Abstract

This research provides an in- depth exploration of recursive chunking methodologies, the mathematical derivation of cosine similarity within the vector space, and a rigorous quantitative evaluation of RAG performance. Our results indicate that NewsMind AI achieves over 90% accuracy in grounding metrics, effectively mitigating hallucinations and providing a superior alternative to traditional lexical search engines and standalone generative models. he information age has changed from time of limited data to world of large information availability, where the large amount of data usually hides important informations where as large language models (LLM) offer advanced reasoning availability, but they face problems like limited real-time knowledge or errors. This paper introduces Newsmind AI, an advanced retrieval augmented generation system (RAG). It is created to provide accurate, time-based and relevant news intelligence. Our system follows a multi-stage process: live news collection through the GNews API, semantic vector creation using the all-mpnet-base-v2 Sentence Transformer, and fast similarity search using the FAISS vector database. By integrating the Groq Llama-3 model as the main generative engine, the system achieves quick response times while maintaining quality and clarity of language. This research explains recursive chunking method and the mathematical concept of cosine similarity in vector space, and a detailed quantitative evaluation of RAG performance. NewsMind AI achieves more than 90% accuracy in reducing hallucination effectively and grounding metrics, providing it as a better alternative to traditional search using keyword and standalone generative models.

Introduction

This research addresses the challenge of information overload in rapidly growing digital news ecosystems, where users struggle to extract meaningful insights from large volumes of constantly updated content. Traditional keyword-based search engines fail to capture semantic meaning, while large language models (LLMs) often produce outdated or hallucinated responses due to reliance on static training data.

To solve this, the study proposes NewsMind AI, a Retrieval-Augmented Generation (RAG) system that combines real-time news retrieval with AI-based reasoning. The system uses external APIs (GNews) to fetch current articles, processes and chunks the text, converts it into vector embeddings using Sentence Transformers, and stores it in a FAISS index for semantic search. Relevant information is then retrieved and passed to a LLaMA-based LLM to generate grounded, context-aware responses.

Key Features

• Real-time news ingestion and processing
• Semantic search using vector embeddings instead of keywords
• RAG framework to reduce hallucinations
• FastAPI backend with Streamlit user interface
• Cosine similarity-based retrieval for relevance matching

Methodology

The system follows a retrieve-then-generate pipeline:

1. User query input
2. Real-time news retrieval
3. Text cleaning and chunking
4. Embedding generation and vector storage
5. Semantic retrieval of top-k relevant chunks
6. LLM-based response generation using grounded context

Results

NewsMind AI significantly outperforms standalone LLMs:

• Accuracy: 92% (vs 64% for LLMs)
• Hallucination rate: <4% (vs 32%)
• Relevance: Very high
  Although response time is slightly higher due to retrieval steps, it improves factual reliability.

Advantages

• Strong reduction in hallucinations
• Real-time, up-to-date news analysis
• Transparent and verifiable outputs
• Efficient semantic retrieval system

Limitations

• Dependence on GNews API quality
• Computational cost of embeddings
• Challenges in handling long multi-article context
• Potential bias from source data

Future Scope

Future improvements include:

• Agentic RAG with iterative search
• Knowledge graph integration
• Multimodal news understanding (text, image, video)
• Multilingual and personalized news summarization

Conclusion

NewsMind AI shows the effectiveness of retrieval augmented generation in improving news usage. By combining modern real-time retrieval with advanced generative reasoning, we have built a system that is accurate and reliable, context-based, and quick. This research confirms that grounding AI in confirmable external data is the most visible path towards building reliable intelligent systems. As the digital information ecosystem grows more complex, tools like NewsMind AI will be useful for turning the noise of the cycle of news into signals of true intelligence.

References

[1] P. Lewis et al., \"Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks,\" in NeurIPS, 2020. [2] N. Reimers and I. Gurevych, \"Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks,\" in EMNLP, 2019. [3] J. Johnson et al., \"Billion-scale similarity search with GPUs,\" IEEE Trans. Big Data, 2019. [4] A. Vaswani et al., \"Attention is All You Need,\" in NIPS, 2017. [5] H. Touvron et al., \"Llama: Open and Efficient Foundation Language Models,\" arXiv, 2023. [6] Groq Inc., \"Real-time AI Inference Whitepaper,\" 2024. [7] M. Douze et al., \"The FAISS Library,\" arXiv, 2024. [8] N. Liu et al., \"Lost in the Middle: How Language Models Use Long Contexts,\" 2023. [9] S. Robertson, \"The Probabilistic Relevance Framework,\" 2009. [10] Meta AI, \"Llama-3 Technical Report,\" 2024. [11] GNews API Documentation, 2025. [12] J. Devlin et al., \"BERT: Pre-training of Deep Bidirectional Transformers,\" 2019.

Copyright

Copyright © 2026 Rohit Tiwari, Harsh Jha, Jiya Kumar, Mrunali Makwana. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.