Crypto Investment Analyst

Multi-Agent AI system that analyzes cryptocurrency data and generates investment reports with actionable recommendations.

Setup Instructions (Mac M1/M2)

1. Create Conda Environment

# Create environment with M1-optimized packages
conda env create -f environment.yml

# Activate environment
conda activate crypto-agent

2. Environment Variables

The file .env.example includes all the API keys that need to be provided for the code to run.

# Copy environment template
cp .env.example .env

# Edit .env with your API keys
nano .env

3. Test Serper Tool

# Test the Serper integration
python src/tools/serper_tool.py

Development Workflow

Adding New Dependencies

As we add more tools, update dependencies:

# Add to pyproject.toml dependencies, then:
pip install -e .

# Export final environment
conda env export > environment-lock.yml

Agent Architecture

The system uses a multi-agent architecture with a supervisor coordinating specialized analysis agents:

System Flow Diagram

                                   USER
                                    |
                         [Investment Question]
                                    |
                                    v
                          ┌─────────────────┐
                          │   SUPERVISOR    │
                          │  (Lead Crypto   │
                          │   Analyst)      │
                          └─────────────────┘
                                    |
                    ┌───────────────┼───────────────┐
                    |               |               |
                    |               |               |
                    v               v               v
           ┌──────────────┐ ┌──────────────┐ ┌──────────────┐
           │ TECHNICAL    │ │ WEB SENTIMENT│ │  CHALLENGER  │
           │ ANALYSIS     │ │   AGENT      │ │   AGENT      │
           │   AGENT      │ │              │ │              │
           └──────────────┘ └──────────────┘ └──────────────┘
                    |               |               |
            [get_multiple_     [serper_search]     [No tools -
             indicators]       [fear_greed_index]   validation only]
                    |               |               |
                    v               v               |
           ┌──────────────┐ ┌──────────────┐       |
           │  TaaPI API   │ │ Serper API   │       |
           │ (Technical   │ │ (News/Web),  │       |
           │ Indicators)  │ │              │       |
           └──────────────┘ └──────────────┘       |
                    |               |               |
                    └───────────────┼───────────────┘
                                    |
                            [Analysis Results]
                                    |
                                    v
                          ┌─────────────────┐
                          │   SUPERVISOR    │
                          │ [Validation &   │
                          │  Synthesis]     │
                          └─────────────────┘
                                    |
                            [Final Report with
                             Buy/Sell/Hold
                             Recommendations]
                                    |
                                    v
                                   USER

Workflow Pattern:
1. User → Supervisor (question + investor profile)
2. Supervisor → Technical Agent → TaaPI API
3. Supervisor → Sentiment Agent → Serper API + Fear/Greed Index
4. Supervisor → Challenger Agent (validate findings)
5. Supervisor → [Re-analyze if needed based on challenger feedback]
6. Supervisor → User (final investment recommendation)

Supervisor Agent (src/agents/supervisor_agent.py)

Lead Crypto Analyst - Orchestrates the analysis workflow and provides final investment recommendations for BTC and ETH.

Workflow:

User-profile is set in user_profiles.json - this is a proxy for current user-holdings and investment style.

1. The supervisor agent takes user-profile and question. It Calls sentiment and technical analysis agents to gather initial data
2. Uses challenger agent to validate findings and identify gaps
3. Re-analyzes based on challenger feedback if needed
4. Provides final buy/sell/hold recommendation with confidence levels
5. For multi-turn conversations, user can ask follow-up questions.

Specialized Agents

• Technical Analysis Agent - Analyzes price data, technical indicators, volume patterns using TaaPI API
• Web Sentiment Agent - Analyzes market sentiment from news, social media, and whale activity using Serper API
• Challenger Agent - Cross-validates findings, flags inconsistencies, ensures thorough evaluation

Multi-Agent Coordination

• LangGraph Supervisor - Routes tasks between agents based on analysis needs
• Memory Checkpointing - Maintains conversation state across multi-turn interactions
• Tool Handoffs - Seamless delegation between specialized agents

Project Structure

src/
├── agents/         # Multi-agent system
│   ├── supervisor_agent.py      # Main coordinator
│   ├── technical_analysis_agent.py
│   ├── web_sentiment_agent.py
│   └── challenger_agent.py
├── tools/          # API integrations
│   ├── serper_tool.py           # News/web search
│   └── taapi_tool.py            # Technical indicators
└── utils/          # Reporting utilities

reports/            # Sample analysis outputs
tests/              # Test suite

Testing & Usage

Run Analysis Demos

# Generate investment reports for different investor types
python main.py

This will generate:

• Single-turn analysis reports for different investor profiles
• Multi-turn conversation demonstrating human-in-the-loop interaction
• Sample outputs saved to reports/ directory

Test Individual Tools

# Test Serper integration
python src/tools/serper_tool.py

# Test TaaPI integration
python src/tools/taapi_tool.py

Sample Outputs

Check reports/ directory for example analysis reports:

• report_aggressive_young.md - High-risk investor analysis
• report_conservative_beginner.md - Conservative strategy
• report_recovery_investor.md - Post-loss recovery approach
• multi_turn_demo.md - Multi-turn conversation example