Copyright © 2026 Authors retain the copyright of this article. This article 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.

@article{199393,
        author = {Prof. Dhananjay Raut and Aryan Balani and Harshala Gaykar and Khushali Gatir and Dhanraj Banglurkar},
        title = {DeepRAG: Implementation and Evaluation of a Multi-Agent Retrieval-Augmented Generation Framework for Autonomous and Verifiable Research Synthesis},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {12006-12015},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=199393},
        abstract = {This paper presents the implementation and experimental evaluation of DeepRAG, a multi-agent Retrieval-Augmented Generation (RAG) frame-work designed for autonomous and verifiable research synthesis. Building upon a previously proposed con-ceptual architecture, this work details the systematic transformation of DeepRAG into a fully functional system, integrating Autogen for multi-agent orches-tration, ChromaDB for semantic vector retrieval, Ser-pAPI for curated web search, Crawl4AI for structured web scraping, and Azure OpenAI for language model-backed synthesis and embedding generation.
The implemented system introduces a parallel sec-tion drafting mechanism coordinated by multiple anal-ysis agents operating concurrently. An automated Quality Evaluation Module validates each generated section against predefined criteria of coherence, rele-vance, and citation completeness prior to its inclusion in the final output. A knowledge base reuse mecha-nism leverages historical query analysis to minimise redundant retrieval operations and accelerate response generation for related research tasks.
Experimental results from representative evaluation runs demonstrate that the system successfully gener-ates structured, multi-section research reports contain-ing an average of 12 sections, drawing on 18 unique web sources processed into 53 indexed knowledge chunks. The scraping module achieved a success rate of ap-proximately 67%, with residual failures compensated by redundant source selection. The parallel agent ar-chitecture and validation-driven pipeline collectively improved both efficiency and output reliability. These findings validate the feasibility of deploying coordi-nated multi-agent RAG systems for real-world research automation and underscore the practical advantages of parallelization, citation traceability, and validation-driven synthesis.},
        keywords = {Multi-agent systems, Retrieval-Augmented Generation, Autonomous research synthesis, ChromaDB, Agent orchestration, Parallel processing, Citation validation, Azure OpenAI, Crawl4AI, SerpAPI.},
        month = {April},
        }