A comprehensive comparison of two leading approaches to building production-ready LLM agents:

• Embabel (Java GOAP): Deterministic planning with enterprise focus
• LangGraph (Python): State-based flexible orchestration

This project demonstrates identical deterministic workflows implemented in two different frameworks for fair side-by-side comparison:

• Pattern: Goal-Oriented Action Planning (GOAP) with deterministic 3-step execution
• Technology: Spring Boot, Spring AI, Java Records
• Focus: Type safety, enterprise integration, high throughput
• Location: java-tool-server/

• Pattern: State graph orchestration with deterministic 3-step execution
• Technology: LangGraph, LangChain, Pydantic
• Focus: Rapid development, flexibility, Python ecosystem
• Location: langgraph-advisor/

Both implementations follow the exact same workflow:

1. Step 1: Call Technical Tool Server → Get quantitative data (SMA, RSI, MACD)
2. Step 2: Call RAG Service → Retrieve qualitative market intelligence
3. Step 3: LLM Synthesis → Generate investment thesis with recommendation

See SIDE_BY_SIDE_COMPARISON.md for detailed execution traces and comparison.

Both frameworks now integrate real Yahoo Finance data:

• Uses yahoofinance-api library (v3.17.0)
• Fetches real historical data (90 days)
• Calculates technical indicators (SMA, RSI, MACD)
• Gracefully degrades to mock data if Yahoo Finance unavailable

• Uses yfinance library (v0.2.66)
• Fetches real market data via Yahoo Finance API
• Calculates same technical indicators
• Multi-layer fallback: Yahoo Finance → MCP Server → Mock data

Note: Yahoo Finance unofficial APIs may hit rate limits. Both implementations handle this gracefully with fallback mechanisms.

curl -X POST http://localhost:8080/ai/agent/thesis \
  -H "Content-Type: application/json" \
  -d '{"ticker":"AAPL"}'

Expected Output:

=== Java GOAP Workflow (Deterministic 3-Step) ===
[Java GOAP - Step 1/3] Call Technical Tool Server
[Java GOAP - Step 1/3] ✓ Retrieved: BULLISH
[Java GOAP - Step 2/3] Call RAG Service
[Java GOAP - Step 2/3] ✓ Retrieved 699 characters
[Java GOAP - Step 3/3] LLM Synthesis
[Java GOAP - Step 3/3] ✓ Recommendation: BUY (Confidence: 75.0%)
=== Java GOAP Workflow Complete ===

python langgraph-advisor/main_advisor.py AAPL

Expected Output:

=== Python LangGraph Workflow (Deterministic 3-Step) ===
[LangGraph - Step 1/3] Call Technical Tool Server
[LangGraph - Step 1/3] ✓ Retrieved: BULLISH
[LangGraph - Step 2/3] Call RAG Service
[LangGraph - Step 2/3] ✓ Retrieved 473 characters
[LangGraph - Step 3/3] LLM Synthesis
[LangGraph - Step 3/3] ✓ Recommendation: BUY (Confidence: 75.0%)
=== Python LangGraph Workflow Complete ===

Quantitative tool server providing financial calculations:

• Technical Indicators: SMA, RSI, MACD calculations
• GOAP Investment Agent: 3-step workflow for investment thesis
• Mock RAG Service: Qualitative market intelligence
• MCP Protocol: HTTP API for microservice communication

State-based orchestration mirroring Java GOAP workflow:

• Node 1: Call Java MCP Tool Server
• Node 2: RAG search for qualitative data
• Node 3: LLM synthesis of investment thesis

Both implementations follow the same 3-step Goal-Oriented Action Planning:

1. Call Quantitative Tool → Get technical indicators
2. Call RAG Service → Retrieve market context
3. LLM Synthesis → Generate investment thesis

cd java-tool-server
mvn spring-boot:run

Server will be available at: http://localhost:8080

python langgraph-advisor/main_advisor.py AAPL

GET http://localhost:8080/ai/tools/health

POST http://localhost:8080/ai/tools/technical-indicators
Body: {"ticker": "AAPL"}

POST http://localhost:8080/ai/agent/thesis
Body: {"ticker": "AAPL"}

public record ThesisResult(
    String ticker,
    String investmentThesis,
    String recommendation,
    double confidenceScore
) {}

class ThesisResult(BaseModel):
    ticker: str
    investmentThesis: str
    recommendation: str
    confidenceScore: float

The mcp.json manifest describes available tools and their schemas, enabling:

• Cross-language tool discovery
• Type-safe remote procedure calls
• Hybrid microservice architectures

Choose Java/Spring AI for:

• Production enterprise systems
• Strict compliance requirements
• High-concurrency microservices
• Integration with existing Java infrastructure

Choose Python/LangGraph for:

• R&D and experimentation
• Data science integration
• Rapid prototyping
• Custom ML model integration

Hybrid Approach (Best of Both):

• Python orchestration calling Java backend tools
• Fast development with stable, scalable foundations