Retrieval-Augmented Generation (RAG) has emerged as a principled approach to grounding large language model (LLM) responses in external knowledge, yet deployed systems face three persistent limitations: single-modality retrieval, fragmented knowledge sources, and monolithic architectures. We propose LinkPulse, an autonomous knowledge-synthesis platform addressing these limitations through three technical contributions: (1) the Tri-Modal Fusion Retriever (TMFR), which dynamically weights dense vector similarity, sparse BM25 keyword matching, and live web-search signals via a lightweight learned gating network; (2) the Multi-Agent Ingestion Pipeline (MAIP), which abstracts heterogeneous sources into a unified vector-indexed knowledge store with provenance tracking; and (3) the Adaptive Context Window (ACW), which combines cross-encoder re-ranking with extractive sentence compression to reduce context dilution before generation. On LinkPulse-Bench, a curated benchmark of 150 queries spanning 1,000+ indexed documents, LinkPulse achieves F1=97.1, outperforming TF-IDF (+13.3), dense-only RAG (+3.4), and Self-RAG (+1.2), with a hallucination rate of 3.2% and mean response latency of 230ms. Code and benchmark are publicly available.
Tapasvi Panchagnula (Fri,) studied this question.