Performance Optimization Guide¶

This guide documents the performance optimizations available in FinWiz's DeepAnalysisCrew for faster, more cost-effective analysis.

Overview¶

FinWiz includes two configurable performance optimizations for the DeepAnalysisCrew:

Phase 1: gpt-4o-mini for Risk Assessment - Use faster, cheaper LLM for risk calculations
Phase 2: Minimal Tool Set - Reduce tool initialization overhead for risk assessor

Both optimizations are enabled by default and can be configured via environment variables.

Configuration¶

Environment Variables¶

Bash

# Phase 1: Use gpt-4o-mini for risk assessment (default: true)
RISK_ASSESSMENT_USE_MINI=true

# Phase 2: Use minimal tool set for risk assessor (default: true)
USE_MINIMAL_RISK_TOOLS=true

Quick Start¶

Default Configuration (Optimized):

Bash

# Both optimizations enabled (fastest, cheapest)
RISK_ASSESSMENT_USE_MINI=true
USE_MINIMAL_RISK_TOOLS=true
uv run python src/finwiz/main.py --ticker AAPL --asset-class stock

Baseline Configuration (No Optimizations):

Bash

# Both optimizations disabled (for comparison)
RISK_ASSESSMENT_USE_MINI=false
USE_MINIMAL_RISK_TOOLS=false
uv run python src/finwiz/main.py --ticker AAPL --asset-class stock

Phase 1: gpt-4o-mini for Risk Assessment¶

Description¶

The risk assessor agent uses gpt-4o-mini instead of the default LLM (GPT-4) for faster, cheaper execution while maintaining accuracy for straightforward risk calculations.

Benefits¶

Faster Execution: gpt-4o-mini is optimized for speed
Lower Cost: Reduced LLM API costs per analysis
Maintained Accuracy: Risk scores remain consistent for standard calculations

Implementation¶

Python

# In src/finwiz/crews/deep_analysis/deep_analysis.py
import os

use_mini = os.getenv("RISK_ASSESSMENT_USE_MINI", "true").lower() == "true"

if use_mini:
    risk_llm = LLM(model="gpt-4o-mini")
    logger.info("Risk assessor using gpt-4o-mini for faster execution")
else:
    risk_llm = self._get_configured_llm()
    logger.info("Risk assessor using default LLM")

When to Disable¶

Complex risk scenarios requiring advanced reasoning
Regulatory compliance requiring specific model versions
Comparative analysis with baseline results
Debugging LLM-related issues

Configuration¶

Bash

# Enable (default)
export RISK_ASSESSMENT_USE_MINI=true

# Disable
export RISK_ASSESSMENT_USE_MINI=false

Phase 2: Minimal Tool Set for Risk Assessor¶

Description¶

The risk assessor uses a minimal, focused tool set instead of the full tool set to reduce initialization overhead and improve startup time.

Benefits¶

Reduced Overhead: Faster tool initialization
Faster Startup: Quicker agent startup time
Focused Tools: Only essential risk calculation tools loaded

Minimal Tool Set¶

The minimal tool set includes only essential tools for risk assessment:

QuantitativeAnalysisTool - Core risk metrics (VaR, CVaR, volatility, Sharpe ratio)
TickerValidationTool - Ticker validation
Asset-Specific Tool - One of:
EnhancedSECAnalysisTool (for stocks)
EnhancedETFAnalysisTool (for ETFs)
EnhancedCryptoAnalysisTool (for crypto)

Implementation¶

Python

# In src/finwiz/crews/deep_analysis/deep_analysis.py
def _get_minimal_risk_tools(self, asset_class: str) -> list:
    """Get minimal tool set for risk assessment only."""
    tools = []

    # Core risk metrics
    tools.append(QuantitativeAnalysisTool(asset_class=asset_class))

    # Ticker validation
    tools.append(TickerValidationTool())

    # Asset-specific tool
    if asset_class == "stock":
        tools.append(EnhancedSECAnalysisTool())
    elif asset_class == "etf":
        tools.append(EnhancedETFAnalysisTool())
    elif asset_class == "crypto":
        tools.append(EnhancedCryptoAnalysisTool())

    return make_tools_robust(tools)

When to Disable¶

Need full tool set for comprehensive analysis
Debugging tool-related issues
Comparative analysis with baseline results
Custom tool requirements

Configuration¶

Bash

# Enable (default)
export USE_MINIMAL_RISK_TOOLS=true

# Disable
export USE_MINIMAL_RISK_TOOLS=false

Performance Testing¶

Test Scenarios¶

Scenario 1: All Optimizations (Fastest)

Bash

RISK_ASSESSMENT_USE_MINI=true USE_MINIMAL_RISK_TOOLS=true \
  time uv run python src/finwiz/main.py --ticker AAPL --asset-class stock

Scenario 2: Phase 1 Only

Bash

RISK_ASSESSMENT_USE_MINI=true USE_MINIMAL_RISK_TOOLS=false \
  time uv run python src/finwiz/main.py --ticker AAPL --asset-class stock

Scenario 3: Phase 2 Only

Bash

RISK_ASSESSMENT_USE_MINI=false USE_MINIMAL_RISK_TOOLS=true \
  time uv run python src/finwiz/main.py --ticker AAPL --asset-class stock

Scenario 4: Baseline (No Optimizations)

Bash

RISK_ASSESSMENT_USE_MINI=false USE_MINIMAL_RISK_TOOLS=false \
  time uv run python src/finwiz/main.py --ticker AAPL --asset-class stock

Expected Results¶

Configuration	Execution Time	Cost	Accuracy
All Optimizations	2-3 min	Low	High
Phase 1 Only	2.5-3.5 min	Low	High
Phase 2 Only	2.5-3.5 min	Medium	High
Baseline	3-4 min	Medium	High

Performance Impact:

20-30% reduction in execution time with all optimizations
Lower LLM API costs with gpt-4o-mini
Maintained accuracy for risk calculations

Validation¶

Risk Score Consistency¶

Compare risk scores between optimized and baseline configurations:

Bash

# Generate optimized results
RISK_ASSESSMENT_USE_MINI=true USE_MINIMAL_RISK_TOOLS=true \
  uv run python src/finwiz/main.py --ticker AAPL --asset-class stock > output_optimized.txt

# Generate baseline results
RISK_ASSESSMENT_USE_MINI=false USE_MINIMAL_RISK_TOOLS=false \
  uv run python src/finwiz/main.py --ticker AAPL --asset-class stock > output_baseline.txt

# Compare risk scores
diff output_optimized.txt output_baseline.txt

Expected Outcome:

Risk scores should be within ±0.1 points
Grades should match (A+, A, B, C, D, F)
Composite scores should be within ±0.05

Quality Assurance¶

Validation Checklist:

Risk scores consistent between configurations
Grades match between configurations
Execution time reduced by 20-30%
No errors or warnings in logs
All tools initialized correctly
API calls successful

Troubleshooting¶

Issue: Risk scores differ significantly¶

Symptoms:

Risk scores differ by >0.2 points between configurations
Grades don't match (e.g., A+ vs A)

Solution:

Check logs for errors or warnings
Verify API keys are valid
Ensure data freshness is consistent
Disable optimizations and compare again

Issue: Performance not improved¶

Symptoms:

Execution time similar between configurations
No cost reduction observed

Solution:

Verify environment variables are set correctly
Check logs for optimization status messages
Ensure gpt-4o-mini is available in your region
Monitor API response times

Issue: Tool initialization errors¶

Symptoms:

Errors during tool initialization
Missing tools in minimal set

Solution:

Disable minimal tools: USE_MINIMAL_RISK_TOOLS=false
Check tool dependencies are installed
Verify asset_class parameter is valid
Review logs for specific error messages

Best Practices¶

Production Deployment¶

Recommended Configuration:

Bash

# Enable all optimizations for production
RISK_ASSESSMENT_USE_MINI=true
USE_MINIMAL_RISK_TOOLS=true

Benefits:

Faster execution for better user experience
Lower costs for high-volume analysis
Maintained accuracy for production use

Development and Testing¶

Recommended Configuration:

Bash

# Use baseline for development and testing
RISK_ASSESSMENT_USE_MINI=false
USE_MINIMAL_RISK_TOOLS=false

Benefits:

Easier debugging with full tool set
Consistent with historical results
Better for comparative analysis

Monitoring¶

Key Metrics to Monitor:

Execution time per ticker
LLM API costs
Risk score consistency
Error rates
Tool initialization time

Logging:

Python

# Logs indicate optimization status
logger.info("⚡ PHASE 2: Using minimal tool set for risk assessor (3 tools)")
logger.info("Risk assessor using gpt-4o-mini for faster execution")

References¶

Implementation Details: See PHASE_1_SPEEDUP_IMPLEMENTATION.md and PHASE_2_SPEEDUP_IMPLEMENTATION.md
Complete Results: See RISK_ASSESSMENT_SPEEDUP_COMPLETE.md
Source Code: src/finwiz/crews/deep_analysis/deep_analysis.py
User Guide: docs/USER_GUIDE.md (Deep Analysis Crew section)
Developer Guide: docs/DEVELOPER_GUIDE.md (Performance Optimization section)

Version: 1.0 Created: 2025-01-25 Last Updated: 2025-01-25 Status: Active