Skip to main content

IJMSRT26JUL026

Page 1

Volume 4/ Issue 7/ July 2026

International Journal of Modern Science and Research Technology ISSN: 2584-2706

A Multi-Agent AI Framework for Automated Research Data Analysis with Blockchain-Based Audit Trail Tajammul Pasha M.Tech Student, Department of Computer Science and Engineering, Ghousia College of Engineering, Ramanagaram, Karnataka-562159, India, Affiliated VTU Belagavi. Dr. Dilshad Begum Professor and Head, Department of Computer Science and Engineering, Ghousia College of Engineering, Ramanagaram, Karnataka-562159, India Dr. Moinuddin S K Assistant Professor, Department of Robotics and Artificial Intelligence, Ghousia College of Engineering, Ramanagaram, Karnataka-562159, India

Abstract Autonomous multi-agent systems built on large language models (LLMs) can carry out entire analytical workflows with little human direction, yet they provide no built-in way to prove what each agent actually did: intermediate reasoning is ephemeral and outputs can be altered without leaving any trace. In this work, a nine-agent autonomous research pipeline implemented using LangGraph and LangChain is presented, which accepts a tabular CSV dataset and a stated research objective and returns statistical analysis, machine-learning models, visualisations and a complete HTML research report. Accountability is attached through a noninvasive instrumentation layer (node_hook.py) that intercepts every agent output, serialises it to a canonical form, computes its keccak256 digest and commits the digest to an append-only smart contract, AgentAuditTrail, on an EVM-compatible blockchain network, producing an immutable and independently verifiable audit trail. Evaluated on five real-world CSV datasets, the pipeline completed every task end to end, achieved an average report-quality rating of 4.12 out of 5.0 from independent reviewers, incurred a blockchain audit-trail overhead of 6.2% of total runtime, and detected 100% of injected modifications across 50 controlled tamper experiments. Novelty in the Research: Existing multi-agent frameworks such as AutoGen, CrewAI and LangGraph provide no tamper-evident record of agent execution, and prior blockchain-provenance systems secure data at rest or in transit rather than the step-by-step behaviour of AI agents. The closest related architecture records agent actions on a private, permissioned chain and requires invasive changes to agent code. In contrast, the present work is, to the best of the authors' knowledge, the first to combine a multi-agent research pipeline with a publicly verifiable blockchain audit trail through a non-invasive hook that requires only two lines of integration code and no modification of any agent, and to quantify the cost of that guarantee (6.2% runtime overhead, about 67,200 gas per action) together with a 100% tamper-detection rate. Keywords: Multi-agent AI, LangGraph, Blockchain, Audit Trail, Data Provenance, Large Language Models, Smart Contracts, Solidity

IJMSRT26JUL026

www.ijmsrt.com

DOI: https://doi.org/10.67317/IJMSRT26JUL026

35


Turn static files into dynamic content formats.

Create a flipbook
IJMSRT26JUL026 by IJMSRT - Issuu