AiDrop

Demo

About

The problem it solves The problem it solves AiDrop-Core is a platform that uses AI to analyze user behavior patterns and preferences within the blockchain ecosystem, enabling personalized marketing and user understanding. Specifically, it solves the following problems: 1. On-chain user behavior data is vast and complex, making it difficult to identify meaningful patterns. AiDrop-Core discovers significant patterns in this data and classifies users through AI models. 2. Blockchain marketing is often untargeted and inefficient. This platform enables segmented targeting based on users' on-chain behavior, greatly improving marketing efficiency. 3. There's a lack of objective criteria for user type classification and behavioral understanding. The project's MBTI-style typing system (with axes like D-T-A-C) systematically categorizes user behavior along four axes, facilitating understanding. 4. Projects lack tools to measure the effectiveness of marketing campaigns in the blockchain space. The marketing tracking module provides metrics and continuous monitoring capabilities to assess campaign performance. 5. Blockchain analytics typically focus on quantitative data rather than qualitative user insights. The AI clustering and classification modules transform raw transaction data into meaningful user profiles.

How it's made

Challenges we ran into 1. Data Synthesis for Testing: Creating representative blockchain user data for model training was challenging. We developed a sophisticated data synthesis approach in test_data/run_ai_models.py that generates realistic transaction patterns mimicking different user types. 2. Feature Engineering Complexity: Extracting meaningful features from raw blockchain data required significant experimentation. We had to identify which transaction attributes (frequency, amounts, contract interactions) were most predictive of user behavior. 3. Multi-dimensional User Classification: Developing a comprehensive typology system that captures diverse blockchain user behaviors was complex. We created a four-axis classification system (Primary Focus, Transaction Pattern, Risk Preference, Community Involvement) to address this. 4. Asynchronous Processing: Handling long-running tasks like image generation within a web API framework required careful implementation of background tasks and state management to prevent timeouts and resource locks. Integration of Multiple ML Models: Combining results from both unsupervised (clustering) and supervised (classification) models required developing a unified data structure and integration approach that preserves insights from both techniques. 5. Cross-module Dependencies: Managing dependencies between the AI, data processing, and API modules while maintaining separation of concerns required careful architecture design and interface definitions.