Stock Analysis Agent

AI-powered multi-agent system delivering institutional-grade, real-time stock analysis using specialised sub-agents for technical, fundamental, sentiment, qualitative, and risk assessment.

Comment

Inspiration

The inspiration for the Stock Analysis Agent came from witnessing the stark divide in financial markets - while institutional investors have access to sophisticated analysis tools and dedicated research teams, individual investors often rely on fragmented information sources and basic analysis. I wanted to democratize access to institutional-grade stock analysis by creating an AI-powered system that could provide comprehensive, multi-dimensional investment insights.

What it does

The Stock Analysis Agent is a comprehensive AI-powered financial analysis system that delivers in-depth stock market analysis through a sophisticated multi-agent architecture. The system consists of:

  • Stock Analysis Coordinator: The root agent that orchestrates the entire analysis process
  • Analysis Orchestrator: Coordinates five specialized sub-agents and synthesizes their outputs
  • Specialized Sub-Agents: Each focuses on a specific domain:
    • Fundamental Agent: Analyzes financial statements, earnings, and company fundamentals
    • Technical Agent: Processes price patterns, charts, and technical indicators
    • Sentiment Agent: Evaluates news, analyst opinions, and market sentiment
    • Qualitative Agent: Assesses management quality, business model, and competitive landscape
    • Risk Assessment Agent: Identifies and quantifies various risk factors

The system produces institutional-grade investment reports that combine all these perspectives into actionable insights for users.

How we built it

We built the Stock Analysis Agent using a hierarchical multi-agent architecture leveraging the Google Agent ADK framework:

Technology Stack:

  • Google Agent ADK as the core framework
  • Gemini 2.5 Pro model for advanced reasoning capabilities
  • Google Search API for real-time data gathering across all sub-agents
  • Python 3.12+ with Poetry for dependency management
  • Vertex AI for cloud deployment and scalability

Architecture Design: The system follows a three-tier approach:

  1. Root Coordination: Stock Analysis Coordinator manages user interactions
  2. Orchestration Layer: Analysis Orchestrator coordinates sub-agents and synthesizes outputs
  3. Specialized Execution: Five domain-specific sub-agents perform targeted analysis

Each sub-agent uses Google Search to gather real-time market data and leverages the same LLM model for consistent reasoning quality.

Challenges we ran into

  1. Multi-Agent Coordination Complexity: Designing a system where multiple specialized agents could work together seamlessly without data conflicts while maintaining consistency across different analysis domains was particularly challenging.

  2. Real-Time Data Integration: Ensuring all sub-agents could efficiently access and process current market data through Google Search API while handling rate limits, data quality variations, and API response times.

  3. Output Synthesis Logic: Creating an intelligent mechanism for the Analysis Orchestrator to combine diverse analysis outputs from technical charts, financial statements, sentiment data, and risk assessments into a coherent, actionable investment report.

  4. Domain Expertise Modeling: Encoding deep financial analysis knowledge into each specialized agent while maintaining the accuracy and sophistication expected from institutional-grade analysis tools.

  5. Performance Optimization: Balancing the comprehensive nature of the analysis with response times, especially when coordinating multiple agents that each need to perform web searches and complex reasoning.

Accomplishments that we're proud of

  • Successfully implemented a modular, hierarchical multi-agent system
  • Achieved institutional-grade analysis quality using only public data and AI
  • Integrated real-time data gathering across all analysis domains
  • Delivered a user-friendly, conversational interface for complex financial insights

What we learned

  • Best practices for multi-agent orchestration and modular AI system design
  • The complexity and value of combining multiple analysis perspectives
  • Effective use of Google Agent ADK and Gemini models for real-world applications
  • Strategies for scalable, maintainable AI architectures

What's next for Stock Analysis Agent

  • Add portfolio analysis using broker MCP and diversification recommendations
  • Implement real-time alerts and monitoring features which can trigger the trades using web hooks events and broker MCP
  • Enable historical backtesting and sector/peer comparisons
  • Integrate ESG and options analysis

Built With

  • agenttool
  • cloud
  • gemini-2.5-pro
  • google
  • google-agent-adk
  • google-search-api
  • poetry
  • python
  • vertex-ai
Share this project:

Updates