International Journal of Scientific Research and Engineering Development

The rapid growth of early-stage startups has created a genuine bottleneck for venture capital analysts, who must process large volumes of heterogeneous financial, strategic, and competitive data before committing to any investment decision. This paper introduces a multi-agent startup investment analysis system that automates the entire research and report-generation pipeline from end to end. The proposed architecture uses a supervisor–worker design orchestrated through LangGraph, where four specialised large language model agents — covering financial analysis, SWOT evaluation, competitor mapping, and investment scoring — operate sequentially on a shared graph state, feeding structured data to a Manager Agent that synthesises a comprehensive investment report. The system draws on Tavily web search and Crunchbase data aggregation for real-time information retrieval, Exa AI neural search for context-aware investor chat, and a React-based progressive web interface backed by a PostgreSQL persistence layer. Testing across a benchmark of 20 publicly known startups shows that the generated reports achieve 87.3% factual coverage and 91.2% citation accuracy when measured against professionally authored analyst reports. The system cuts manual research time by an estimated 78% and produces consistently structured output across diverse industry sectors. These findings suggest that LLM-orchestrated multi-agent pipelines offer a practical, scalable path toward democratising professional-grade startup due diligence.

Chandan C R, Likhitha T R, Pavan Tej, Susma N R,"A Multi-Agent Framework for Automated Startup Investment Analysis Using Large Language Models and Knowledge-Graph Orchestration" International Journal of Scientific Research and Engineering Development, V9(3): Page(365-371) May-June 2026. ISSN: 2581-7175. www.ijsred.com. Published by Scientific and Academic Research Publishing.