The Classification Engine
The Classification Engine is a specialized agent responsible for extracting and categorizing crucial data from incoming Twitter and Telegram messages. Its core functions include: Narrative Detection- Determines overarching themes based on context.
- Examples: DeFi, AI Agents, DeFAI, Lending, etc. Entity Extraction
- Identifies references to crypto projects.
- Example: Recognizing “@StrawberryAI_5” as BERRY using an internal lookup table. Ticker Extraction
- Detects and infers crypto tickers such as BERRY, BTC, even if they are not explicitly mentioned, leveraging similarity search and contextual analysis. Sentiment Analysis
- Assigns sentiment scores (0 = bearish, 10 = bullish) to identified narratives, projects, and tickers. Each sentiment score includes a confidence level (0 to 1, with decimal precision). All extracted data is then processed by the Big Data Engine for deeper analysis.
The Big Data Engine
The Big Data Engine serves as Luigi’s analytical core, processing vast amounts of time-series data to stay ahead of market trends. In crypto, narratives and market buzz evolve rapidly, requiring real-time adaptation. The Big Data Engine utilizes a proprietary algorithm that updates rankings every second to reflect emerging market dynamics. It optimizes for key factors, including: Novelty Detection – Identifying new and emerging trends. Trend Analysis – Recognizing momentum shifts. Narrative Rankings – Prioritizing dominant themes. Ticker Rankings – Assessing sentiment, correlation, and trend strength. By continuously refining these metrics, the Big Data Engine ensures that Luigi delivers insights based on the freshest and most relevant market intelligence.The Reasoning Engine
The Reasoning Engine is responsible for final decision-making, distilling large datasets into actionable recommendations.- Every hour, it ingests the latest snapshot from the Big Data Engine, processing over 100 ranked tickers and more than 2 million tokens per run.
- To handle this volume, it operates through a recursive agent loop:
- For each ranked ticker, it performs an in-depth analysis to generate a “worth buying” score (0 to 10).
- The top 10 scoring tickers proceed to the final evaluation phase.
- This phase incorporates real-time market data, sentiment analysis, social chatter, and external web-sourced insights.
- Ultimately, the process results in 1 to 3 recommended picks per hour.