Year: 2025
Type: article
Abstract: In recent years, Large Language Models (LLMs) have transformed the landscape of intelligent systems, yet their standalone capabilities remain constrained by limited context understanding and static knowledge retrieval.This study presents a comprehensive review of LLMs with a focus on designing a Multi-Agent LLM architecture that fosters collaborative reasoning, decentralized knowledge access, and dynamic task delegation.We introduce a novel hybrid search framework that combines dense semantic retrieval and sparse keyword matching to optimize contextual grounding in real-time, also use vector databases for more fast and optimal results.Leveraging vector databases such as AstraDB, FAISS and Chroma, along with embedding models from OpenAI and HuggingFace, our system encodes documents and user queries into rich vector representations.These vectors support both cosine similarity-based vector search and traditional TF-IDF methods, allowing parallelized semantic and lexical search.The outputs are unified, ranked, and synthesized using prompt templates within an LLM, enabling accurate, explainable, and adaptive responses.This blueprint lays the groundwork for a scalable and modular intelligence system wherein agents operate asynchronously yet cooperatively, driven by shared objectives and enriched by contextual memory.Our work not only bridges the gap between retrieval and generation but also paves the way for deployable multi-agent intelligence frameworks.
Language: English
Topic: Topic Modeling
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: hybrid
OpenAlex ID: W4411771462