COO Assistant - Orkes Agent

An AI-powered COO Assistant that orchestrates multiple agents to analyze business data and provide actionable recommendations with continuous learning capabilities.

Features

• Parallel Analysis: Runs multiple analysis agents simultaneously
• Data Synthesis: Combines insights from reviews and metrics
• Smart Recommendations: Generates prioritized, actionable recommendations
• Continuous Learning: Tracks outcomes and improves over time
• Demo Mode: Simulates the learning process for demonstrations

Setup

1. Get Orkes Credentials

   • Sign up at https://developer.orkescloud.com/ for free
   • Or get credentials from the Orkes booth at the hackathon
   • You'll need both API Key and Secret Key

2. Configure Environment

   cp .env.example .env
   # Edit .env and add:
   # - ORKES_API_KEY (your API key ID)
   # - ORKES_SECRET_KEY (your API secret)

3. Install Dependencies

   npm install

Running the Agent

Demo Mode (No API Key Required for Testing)

# Quick demo - single analysis
npm run demo

# Continuous improvement demo - shows learning over time
npm run demo:continuous

API Server Mode (Recommended)

# Start the API server - listens for events
npm run server

# The server will start on port 3000 (or PORT env variable)
# API endpoints:
# - POST /api/analyze - Trigger analysis
# - POST /api/webhook/:source - Receive webhooks
# - GET /api/metrics - Get learning metrics
# - GET /api/recommendations - Get recommendation history
# - GET /api/queue/status - Check event queue status

Production Mode (Single Run)

npm start

Architecture

The agent consists of:

1. Workflow Definition (src/workflows/coo-assistant-workflow.js)

   • Defines the orchestration flow
   • Manages parallel task execution

2. Workers (src/workers/)

   • DataCollectorWorker: Gathers data from sources
   • ReviewAnalyzerWorker: Analyzes customer reviews
   • MetricsAnalyzerWorker: Analyzes product metrics
   • SynthesisWorker: Combines insights
   • RecommendationWorker: Generates recommendations

3. Learning Coordinator (src/learning/learning-coordinator.js)

   • Tracks recommendations and outcomes
   • Calculates improvement metrics
   • Manages continuous learning loop

Integration Points

The agent is designed to integrate with:

• Bright Data: For review scraping (Engineer 2)
• Mixpanel: For product analytics (Engineer 3)
• Dashboard: For displaying recommendations (Engineer 3)

API Usage

When running in production mode, trigger analysis via API:

const result = await conductorClient.workflowResource.startWorkflow({
  name: 'coo_assistant_analysis',
  version: 1,
  input: {
    startup_id: 'your_startup_id'
  }
});

Workflow Visualization

The workflow executes in this order:

1. Collect Data (from multiple sources)
2. Parallel Analysis (reviews + metrics)
3. Synthesize Insights
4. Generate Recommendations
5. Track for Learning

Success Metrics

The agent tracks:

• Recommendation success rate
• Improvement trends
• Confidence levels
• Time to impact