Abstract

PortfoliAI is an AI-powered portfolio assistant that analyzes users’ investment portfolios, evaluates risks, conducts research, and provides actionable recommendations. Leveraging a multi-agent system, it integrates LLMs via the Gemini API, custom web search tools using Google Programmable Search, and automated CSV/PDF data ingestion. The system demonstrates scalable, asynchronous agent collaboration to deliver structured insights with minimal user input.

Introduction

Investors often face challenges in tracking portfolio performance, understanding risks, and gathering relevant market research. PortfoliAI addresses these challenges by employing a multi-agent architecture, where specialized agents handle portfolio analysis, risk assessment, research, and recommendation generation. Users can upload CSV or PDF files, and the system converts them into structured data for analysis.

Key motivations:

• Provide actionable insights in real-time.
• Automate risk evaluation and research for better decision-making.
• Demonstrate the effectiveness of agent orchestration frameworks (LangGraph) in financial applications.

Methodology

PortfoliAI employs a modular, asynchronous multi-agent workflow:

1. Input Conversion: The InputConverter agent processes CSV or PDF files to extract portfolio tables in a structured format.
2. Portfolio Analysis: PortfolioAgent analyzes the holdings, valuations, and performance.
3. Risk Assessment: RiskAgent evaluates potential risks and volatility.
4. Research Agents: Two ResearchAgents gather market insights on portfolio and risks using a custom Google search tool.
5. Recommendation Generation: RecommendationAgent combines outputs from other agents and provides actionable suggestions.

Orchestration & Tools Used:

• LangGraph: Manages agent workflow and state propagation.
• Gemini API: Large language model for advanced analysis.
• Google Programmable Search Tool: Custom search for market and risk research.
• Python Libraries: Pandas, PDFPlumber, Streamlit for frontend visualization.

Architecture Diagram:
Users upload CSV/PDF → InputConverter → PortfolioAgent → RiskAgent → ResearchAgents → RecommendationAgent → Output

Experiments

PortfoliAI was tested with sample portfolios in CSV and PDF formats. Key observations:

User Interaction:

1. Upload CSV or PDF → wait for processing → results displayed in sections: Portfolio Analysis, Risk, Research, Recommendations.
2. Outputs are rendered in Markdown tables and textual summaries for clarity.

Conclusion

PortfoliAI demonstrates how multi-agent systems can effectively automate portfolio analysis and decision-making. By integrating LLMs, orchestration frameworks, and custom search tools, the system provides real-time insights and actionable recommendations, reducing the manual effort required by users. This project highlights the potential of agentic AI in financial technology applications.

GitHub Repository: https://github.com/shanks1554/portfoliai.git
Live Demo: https://portfoliai.onrender.com/