FinAgent: Multi-Agent Financial Forecasting

Inspiration

We were inspired by the complexity and inefficiency of traditional portfolio management. Most investors rely on spreadsheets, static dashboards, or intuition, which can miss nuanced trends in stock behavior. I wanted to create a system that combines AI decision-making, Modern Portfolio Theory (MPT), and interactive visualization to provide actionable, precise, and visually clear investment insights.

What It Does

This project generates AI-driven portfolio recommendations for FAANG + NVIDIA stocks, combining MPT-based allocations with decision engine suggestions. It calculates volatility, confidence, optimal buy dates, and exact holding periods, and visualizes all data in professional interactive charts. Users can compare AI vs MPT allocations, evaluate risk, and view tactical, actionable investment recommendations.

How We Built It

Collected historical stock price data using Yahoo Finance (yfinance).

Calculated returns, volatility, seasonality, and optimal holding periods for each stock.

Built a decision engine agent that recommends buy/hold actions, AI allocations, and best entry dates.

Created interactive visualizations using Plotly:

Portfolio allocation (AI vs MPT)

Confidence vs volatility scatter

Tactical timeline for exact buy dates and holding periods

Exported visualizations as interactive HTML files, enabling production-ready sharing or embedding.

Challenges We Ran Into

Integrating AI recommendations with MPT allocations without producing conflicting or misleading suggestions.

Displaying exact holding periods on a timeline while maintaining clear color coding and hover info.

Ensuring production-grade interactive visualizations without relying on heavy dashboard frameworks or APIs.

Handling data gaps, missing values, and noisy historical stock data for accurate calculations.

Accomplishments That We’re Proud Of

Successfully built a multi-agent system that outputs actionable investment recommendations with exact holding periods.

Designed professional, interactive, and visually appealing charts without using complex dashboard frameworks.

Created a workflow that combines finance theory, AI, and visualization in a single cohesive tool.

Demonstrated the ability to automate data analysis and tactical recommendations for real-world investment decisions.

What We Learned

How to blend quantitative finance (MPT) with AI-driven decision-making for actionable insights.

Techniques for calculating risk-adjusted metrics, optimal entry points, and exact holding periods.

Best practices for interactive financial visualization that is intuitive and production-ready.

How to design a reusable multi-agent architecture for generating structured insights and visual outputs.

What’s Next for Multi-Agents for Financial Predictions

Introduce reinforcement learning agents to dynamically adjust portfolio allocations in real-time.

Expand coverage to multi-sector, global portfolios beyond FAANG + NVIDIA.

Integrate real-time market data and predictive signals for live recommendations.

Add automated alerting agents that notify users when conditions match tactical entry or exit strategies.

Explore risk mitigation strategies and scenario simulations to further optimize portfolio performance.

Built With

• agents
• ai
• claude
• llm
• openai
• yahoo