AlphaMesh: a personalized stock investment app powered by multi-agent LLM orchestration

Lim, Yeu Chuan (2026) AlphaMesh: a personalized stock investment app powered by multi-agent LLM orchestration. Final Year Project, UTAR.

PDF Download (4Mb)

Abstract

The increasing complexity of financial markets and the limitations of existing investment platforms present significant challenges for retail investors, who often face information overload, limited personalized guidance, and opaque analytical tools that are difficult to interpret. Traditional robo-advisors usually rely on static questionnaire-based profiling, while modern investment applications often overwhelm users with large amounts of market data, charts, and financial indicators without sufficient explanation or personalization. This project presents AlphaMesh, a personalized stock investment application powered by a multi-agent Large Language Model (LLM) orchestration framework that aims to provide more explainable, adaptive, and user-oriented investment support. AlphaMesh was developed as a working web-based prototype using a FastAPI backend and Vite frontend, with interface design assistance from Google Stitch AI. The system consists of an Orchestrator Agent, News Analysis Agent, and Fundamental Analysis Agent, which work together to interpret user queries, retrieve relevant user and portfolio context, analyse recent financial news, compute fundamental financial metrics, and synthesize personalized investment responses. To support long-term personalization, AlphaMesh implements a dual-level memory architecture that combines ChromaDB vector retrieval for contextual article recall with Neo4j graph-based memory for structured financial relationships and user-specific interest tracking. The system also supports portfolio management, conversation history, and dynamic visualization to help users understand market trends and company performance more clearly. Evaluation results show that AlphaMesh achieved an average score of 7.38 out of 10 in scenario-based response quality testing using LLM-based judge, successfully captured evolving user preferences across conversations, and demonstrated richer evidence retrieval through its dual-store RAG pipeline. Overall, AlphaMesh demonstrates the potential of multi-agent LLM systems to create a more transparent, personalized, and accessible investment analysis tool for both beginner and experienced retail investors alike in Malaysia.

Actions (login required)