How each of the 10 hackathon sponsor tools is utilized in the Autonomous Global Event Arbitrage Agent's real-time trading pipeline.
Senso acts as the agent's working memory, persisting context between cycles. It tracks which workflow step the agent is in (ingesting, analyzing, predicting, executing, learning), stores the latest opportunity scores, and maintains the integration status map.
Airbyte feeds the agent with external economic indicators โ CPI, unemployment rate, Fed funds rate, GDP growth โ that influence market behavior. These macro signals are combined with Robinhood price data and Tavily sentiment to build a complete market picture.
Tavily enriches the agent's market view with real-time web sentiment. Each cycle, it searches for market momentum signals and trending financial news to gauge whether sentiment supports the detected arbitrage spread. The sentiment score directly influences Fastino's prediction.
Reka's Vision API analyzes candlestick charts to detect non-textual patterns like bullish engulfing, double bottoms, and volume spikes. These visual signals are routed through Yutori to Neo4j for historical correlation matching โ adding a dimension of analysis beyond pure numerical data.
Neo4j stores every market event, price snapshot, and trade outcome as nodes in a graph. It finds historical correlations โ like "BTC rises +3.2% within 48h of a dovish Fed signal" โ that inform the Fastino prediction model. Trade outcomes update the graph, creating the agent's learning loop.
Fastino's fine-tuned model evaluates each arbitrage opportunity using spread data, Tavily sentiment, Reka visual patterns, and Neo4j historical correlations โ all managed by Senso context. It outputs a 0โ1 probability score. Scores โฅ0.75 trigger trades; โฅ0.95 trigger anomaly alerts.
Yutori's N1 Navigator routes data between pipeline stages based on type and priority. Market data goes to Neo4j (high priority), sentiment to Fastino (medium), trade results to Numeric (high). It also makes the final execute/skip decision based on the predicted score threshold.
Numeric provides the clean P&L data that drives the agent's learning loop. Every trade is logged to a structured ledger with entry/exit prices, quantity, and P&L. The resulting win rate and Sharpe ratio feed back into Neo4j to improve future Fastino predictions.
When the agent detects a highly unusual, high-risk/high-reward anomaly (score โฅ0.95 or spread >5%), Modulate synthesizes an urgent voice warning. This provides an audible alert layer beyond the visual dashboard โ critical for time-sensitive trading anomalies.
Render hosts the continuously running agent as a Docker container. The render.yaml
Blueprint configures both a Web Service (FastAPI dashboard + API) and a Background Worker (the
autonomous agent loop), satisfying the 2+ Render services requirement. Deployed from GitHub with
auto-deploy on push.