5. Utilization of G.A.M.E Framework

FriendFi achieves advanced credit scoring by utilizing Virtuals Protocol's G.A.M.E Framework. The G.A.M.E Framework is a framework for AI agents to make decisions autonomously, and functions as the "brain" of AI agents in FriendFi.

• Overview of G.A.M.E Framework: G.A.M.E (Generative Autonomous Multimodal Entities) is a modular framework that enables agents to autonomously plan and make decisions. It is a decision-making engine built on foundation models and can be used to power agents in various environments and platforms.

• Role of G.A.M.E Framework in FriendFi: In FriendFi, the G.A.M.E Framework plays a crucial role primarily in credit scoring. By using the G.A.M.E Framework, AI agents can analyze user behavior data and evaluate their creditworthiness.

• Utilization of G.A.M.E Framework in Credit Scoring:

  • High-Level Planner (HLP) and Low-Level Planner (LLP): The G.A.M.E Framework consists of two main components: HLP and LLP. HLP determines long-term plans, while LLP determines specific actions.

  • Role of HLP in Credit Scoring: HLP comprehensively considers activities on EchoSphere, on-chain activities, and credit token holdings to evaluate a user's long-term creditworthiness and sets a target credit score.

    • Example: Target Credit Score = f(EchoSphere Activity, On-chain Activity, Credit Token Holdings)

  • Role of LLP in Credit Scoring: LLP determines specific actions to achieve the target credit score set by HLP. For example, to improve a user's credit score, LLP may execute the following functions:

    • EchoSphere Activity Analysis Function: Analyzes the user's statements, frequency, engagement, etc.

      • Statement Score = g(Quality of Statement, Frequency of Statement, Engagement Rate)

    • On-chain Activity Analysis Function: Analyzes the user's transaction history, asset holdings, etc.

      • Transaction Score = h(Transaction Frequency, Transaction Volume, Asset Holdings)

    • Credit Token Evaluation Function: Analyzes the user's credit token holdings, transaction history, etc.

      • Token Score = i(Holdings, Transaction Volume, Holding Period)

  • Feedback Loop: The results of actions executed by LLP are fed back to HLP, which helps adjust the target credit score and formulate new action plans.

    • Adjusted Target Credit Score = Target Credit Score + α(Statement Score, Transaction Score, Token Score)

    • Here, α represents the weighting coefficient for each score.

  • Adding Custom Functions: In FriendFi, developers can define their own functions and add them to LLP. This enables more precise credit scoring specialized for specific use cases.

  • Example of Function Definition (Credit Token Price Prediction Function):

    {
      "fn_name": "predict_credit_token_price",
        "fn_description": "Predicts the future price of a credit token by analyzing past price data and market trends.",
        "args": [
          {
            "name": "token_symbol",
            "description": "Symbol of the token for which to predict the price",
            "type": "string"
          },
          {
            "name": "timeframe",
            "description": "Period for which to predict (e.g., 1 day, 1 week, 1 month)",
            "type": "string"
          }
        ],
      "config": {
        "method": "post", 
        "url": "<https://api.example.com/predict>", 
        "headers": {
          "Authorization": "Bearer your_api_key"
        },
        "payload": {
          "model": "price_prediction_model",
          "data": {
              "token_symbol": "{{token_symbol}}",
              "timeframe": "{{timeframe}}"
            }
        },
        "success_feedback": "Predicted Price: {{#response.prediction}} {{value}} {{/response.prediction}}",
        "error_feedback": "Price prediction error"
      }
    }