Abstract: In the field of digital forensics, the escalating volume and complexity of evidence data constitutes a critical bottleneck for investigators. Traditional manual extraction and analysis workflows are time-intensive and error-prone, particularly when dealing with large-scale datasets from seized mobile and computing devices. This paper presents an innovative AI-driven conversational system designed to substantially accelerate forensic analysis pipelines. Leveraging a Retrieval-Augmented Generation (RAG) architecture, the system ingests and indexes data encoded in the Universal Forensic Data Representation (UFDR) format, enabling forensic analysts to interrogate complex digital evidence through natural language queries. Our proposed architecture integrates a transformer-based embedding model, a vector similarity retrieval engine, and a large language model (LLM) generation layer to deliver contextually accurate, evidence-grounded responses. Experimental evaluations on a curated UFDR dataset demonstrate that our system reduces mean query-response time by 67% compared to conventional keyword-based tools, achieving a retrieval precision of 0.91 and an answer faithfulness score of 0.88. These results validate the efficacy of RAG-based conversational interfaces for investigative workflows and signal a paradigm shift in digital forensic methodology.

Keywords: Digital forensics, retrieval-augmented generation, large language models, UFDR, natural language processing, conversational AI, evidence retrieval, transformer models.

Downloads: | DOI: 10.17148/IJIREEICE.2026.14433