Widespread deployment of large language models (LLMs) across knowledge-intensive industries has brought their core architectural weakness into sharp focus: a fixed internal knowledge state that cannot reflect post-training developments and that carries a persistent risk of generating plausible yet factually unsupported content. Retrieval-Augmented Generation (RAG) offers a compelling remedy by coupling the generative capacity of LLMs with a dynamically queryable external knowledge store, thereby decoupling reasoning from memorisation. This work conducts a structured systematic review of RAG research spanning the period 2020 through 2026, charting its progression from rudimentary retrieve-then-read configurations toward sophisticated pipelines that incorporate modular retrieval components and autonomous agent-driven reasoning. Core technical mechanisms are analysed in depth, covering bi-encoder and late-interaction retrieval models, multi-passage fusion strategies, and the complementary roles of lexical and semantic search. Quantitative evidence drawn from widely adopted open-domain benchmarks confirms that retrieval-augmented systems consistently surpass purely parametric baselines on factual question-answering tasks. The review further examines how self-critique loops and structured knowledge graphs are being employed to reduce model hallucinations at scale. Concluding observations chart priority research directions in multimodal retrieval, temporal knowledge decay, and privacy-safe retrieval, positioning RAG as the foundational knowledge infrastructure for next-generation trustworthy AI deployments.

Retrieval-Augmented Generation has undergone a transformation from a conceptually attractive but architecturally simple prototype into a mature, industrially deployable paradigm for knowledge-grounded AI. By externalising the knowledge store and introducing structured retrieval between user intent and model generation, RAG directly addresses the two most consequential weaknesses of pre-trained LLMs: temporal knowledge decay and hallucination under uncertainty. The review presented here documents a clear developmental arc: from single-stage retrieve-and-read to modular pipelines with hybrid retrieval, adaptive re-ranking, and self-reflective generation; and from static document indices to dynamic graph-structured knowledge bases navigated by autonomous reasoning agents. Benchmark trajectories confirm that each architectural refinement delivers measurable performance gains, particularly on complex multi-hop tasks that require synthesising evidence distributed across many documents. For the engineering practitioner, RAG’s modular design is its greatest practical asset: domain knowledge can be updated, audited, and replaced independently of the generator, enabling compliance with data governance requirements that would be impractical to satisfy through model fine-tuning.

[1] P. Lewis et al., “Retrieval-augmented generation for knowledge-intensive NLP tasks,” in Proc. NeurIPS, 2020. [2] K. Guu, K. Lee, Z. Tung, P. Pasupat, and M. W. Chang, “REALM: Retrieval-augmented language model pre-training,” in Proc. ICML, 2020. [3] J. Gao et al., “Retrieval-augmented generation for large language models: A survey,” arXiv:2312.10997, 2023. [4] W. Yu et al., “Survey of retrieval-augmented generation: Architectures, techniques and applications,” IEEE Trans. Knowl. Data Eng., 2024. [5] V. Karpukhin et al., “Dense passage retrieval for open-domain question answering,” in Proc. EMNLP, 2020, pp. 6769–6781. [6] O. Khattab and M. Zaharia, “ColBERT: Efficient and effective passage search via contextualized late interaction over BERT,” in Proc. SIGIR, 2020, pp. 39–48. [7] G. Izacard and E. Grave, “Leveraging passage retrieval with generative models for open domain question answering,” in Proc. EACL, 2021, pp. 874–880. [8] S. Borgeaud et al., “Improving language models by retrieving from trillions of tokens,” in Proc. ICML, 2022. [9] W. Shi et al., “REPLUG: Retrieval-augmented black-box language models,” arXiv:2301.12652, 2023. [10] A. Asai et al., “Self-RAG: Learning to retrieve, generate, and critique through self-reflection,” in Proc. ICLR, 2024. [11] H. Trivedi et al., “Interleaving retrieval with chain-of-thought reasoning for knowledge-intensive multi-step questions,” in Proc. ACL, 2023, pp. 10014–10037. [12] S. Es et al., “RAGAs: Automated evaluation of retrieval augmented generation,” in Proc. EACL System Demonstrations, 2024, pp. 150–158. [13] P. Sarthi et al., “RAPTOR: Recursive abstractive processing for tree-organized retrieval,” in Proc. ICLR, 2024. [14] B. Peng et al., “Graph retrieval-augmented generation: A survey,” J. ACM, vol. 37, no. 4, Art. 111, Sep. 2024. [15] S. Yao et al., “ReAct: Synergizing reasoning and acting in language models,” in Proc. ICLR, 2023. [16] B. Edge et al., “GraphRAG: A graph-based RAG approach for global sensemaking,” Microsoft Research, 2024. [17] Q. Zhao et al., “LongRAG: A dual-perspective retrieval-augmented generation paradigm,” in Proc. EMNLP, 2024, pp. 22600–22632. [18] Z. Jiang, X. Ma, and W. Chen, “LongRAG: Enhancing retrieval-augmented generation with long-context LLMs,” arXiv:2406.15319, 2024. [19] S. Wang et al., “InstructRetro: Instruction tuning post retrieval-augmented pretraining,” arXiv:2310.07713, 2023. [20] RAG 2.0: The 2025 guide to advanced retrieval-augmented generation. [Online].Available: https://vatsalshah.in/blog/the-best-2025-guide-to-rag

© 2025 International Journal of Engineering and Techniques (IJET).

Digital Object Identifier (DOI)

IJET assigns a unique DOI to every accepted article via Zenodo for persistent linking and citation. Your article’s DOI: https://doi.org/{{doi}}

Open DOI About Zenodo DOI

Close