OWL (Optimized Workforce Learning)

OWL (Optimized Workforce Learning) is a state-of-the-art multi-agent collaboration framework built on top of the CAMEL-AI framework. It is designed to automate real-world tasks through dynamic agent interactions. OWL achieved a score of 58.18 on the GAIA benchmark, ranking #1 among open-source frameworks.

Core Capabilities:

• Online Search: Real-time information retrieval via Wikipedia, Google, DuckDuckGo, Baidu, Bocha, etc.
• Multimodal Processing: Handles internet or local video, image, and audio data.
• Browser Automation: Uses Playwright to simulate user actions like scrolling, clicking, typing, and downloading.
• Document Analysis: Extracts content from Word, Excel, PDF, and PowerPoint, converting them to text or Markdown.
• Code Execution: Executes Python code via an interpreter.
• Extensive Toolkits: Includes Model Context Protocol (MCP) support and specialized toolkits (e.g., Arxiv, GitHub, Google Maps, Search, Browser, etc.).

The Interview Preparation Assistant is a multi-agent AI system built on the OWL framework designed to automate job interview preparation. It uses specialized agents to research companies, generate tailored questions, and create structured study plans. This system is intended to replace fragmented, manual research with a cohesive, personalized preparation package.

Core Agent Roles

• Company Research Agent: Conducts real-time research on company culture, technical stacks, product offerings, and interview processes using web search and content processing.
• Question Generation Agent: Produces technical and behavioral questions based on the specific job role, required skills, and company-specific methodologies.
• Preparation Plan Agent: Organizes findings into actionable schedules, including day-by-day study topics, mock interview scenarios, and technical practice problems with solutions.

Technical Stack

• Framework: OWL Multi-Agent Framework
• Tools: Dynamic Research Tools (real-time web search)
• Interface: Streamlit
• Models: Advanced LLMs (OpenAI/Gemini)

The A-Share Investment Agent is a multi-agent system built on the OWL and Camel (v0.2.36) frameworks. It uses 10 specialized agents organized into three functional layers to analyze the Chinese A-share market:

1. Data & Analysis Layer

• Market Data Agent: Collects and preprocesses market data.
• Technical Analyst: Analyzes technical indicators and generates signals.
• Fundamentals Analyst: Analyzes fundamental data and generates signals.
• Sentiment Analyst: Analyzes market sentiment and generates signals.
• Valuation Analyst: Calculates intrinsic stock value and generates signals.

2. Research & Debate Layer

• Researcher Bull: Provides bullish perspectives.
• Researcher Bear: Provides bearish perspectives.
• Debate Room: Integrates bull and bear views into a final consensus.

3. Decision Layer

• Risk Manager: Calculates risk metrics and sets position limits.
• Portfolio Manager: Formulates final trading decisions and generates orders.

The system follows a multi-agent pipeline to reach a final investment decision:

1. Data Collection: The Market Data Agent collects A-share market data and news via the akshare API.
2. Multi-dimensional Analysis: Four specialized agents perform independent analyses:
   • Technical Analyst: Analyzes technical indicators (e.g., Moving Averages, RSI, MACD, Bollinger Bands).
   • Fundamental Analyst: Evaluates financial health (e.g., Net Profit, Revenue, Debt Ratio).
   • Sentiment Analyst: Analyzes market sentiment from news and social media.
   • Valuation Analyst: Performs valuation using models like DCF and relative valuation.
3. Research & Debate:
   • Bull Researcher: Provides a bullish perspective based on the analysis.
   • Bear Researcher: Provides a bearish perspective based on the analysis.
   • Debate Room: Integrates the bull and bear viewpoints to form a balanced opinion.
4. Risk Assessment: The Risk Management Agent evaluates volatility and sets trading limits (e.g., max position size, max drawdown).
5. Final Decision: The Portfolio Management Agent synthesizes all signals, debate results, and risk constraints to generate a final action (e.g., hold, buy, sell).

The Stock Analysis Agent is a multi-agent system built on the OWL framework that generates comprehensive investment reports. It consists of several specialized components:

• Stock Analysis Agent (Main Agent): A RolePlaying Agent (from Camel-ai) that orchestrates the analysis process.
• SEC Agent: A ChatAgent that uses the SEC Tool to retrieve and summarize company financial data. It is recommended to use lightweight models like Zhipu's GLM-4-Flash here to manage token costs, as financial statements can be extremely large.
• Stock Analysis Tools:
  • Search Tool: Uses search engines (like Baidu) to collect basic company information.
  • SEC Tool: Retrieves company basic information and financial statements via sec-api.
  • Report Write Tool: A file editing tool that writes the final investment analysis report to a .md file.

Workflow: The main agent coordinates with the SEC Agent to gather data, uses search tools for context, and finally uses the write tool to generate the report.

The investment agent system uses a multi-agent architecture to produce a comprehensive investment decision. The output consists of several specialized agent logs that contribute to a final decision:

1. Market Data Agent: Analyzes trends and provides support/resistance levels.
2. Technical Analyst: Evaluates technical indicators (Moving Averages, MACD, RSI, Bollinger Bands) to provide a signal (e.g., bearish, bullish).
3. Fundamentals Analyst: Examines financial health, profitability, and valuation ratios (P/E, P/B).
4. Sentiment Analyst: Scans news and social media to determine market mood.
5. Valuation Analyst: Performs intrinsic value calculations (e.g., DCF model).
6. Researcher (Bull & Bear): These agents debate the findings, representing opposing viewpoints.
7. Debate Room: Synthesizes the Bull and Bear arguments into a neutral or consensus signal.
8. Risk Manager: Calculates risk scores, volatility, and suggested position sizes.
9. Portfolio Manager: The final decision-maker that outputs the action (e.g., hold, buy, sell), quantity, and confidence level.

Each agent typically returns a JSON object containing a signal, confidence, and reasoning field.