Tuğberk AkbulutGet in touch
Back to work

Racfathers AI Platform

2025

Two LangGraph agents running crypto market analysis and trading psychology in production, across five surfaces.

PythonTypeScriptLangGraphFastAPIpgvector

As Software Developer at Racfathers, I led architecture and delivery of a multi-agent AI platform for crypto market analysis and trading psychology, deployed across Telegram, WhatsApp, Farcaster, Discord, and a REST API.

Telegram WhatsApp Farcaster Discord REST API Consigliere Crypto market analysis Racfella Trading psychology Shared pgvector memory User profiling + RAG retrieval

Two LangGraph agents anchor the system: Consigliere for crypto market analysis, and Racfella for WhatsApp-based trading psychology coaching. Both share a pgvector-backed semantic memory layer with user profiling and RAG retrieval, and exchange credentials are secured with AES-256-GCM envelope encryption plus single-use onboarding tokens rather than long-lived stored secrets.

I also owned the full ML lifecycle for two production models feeding the platform’s risk signals: an LSTM for Bitcoin price forecasting and an XGBoost classifier for rugpull detection, covering feature engineering through Dockerized deployment, alongside Base and Ethereum integrations with millisecond-level Uniswap V2 liquidity monitoring via Alchemy WebSocket streaming.

Racfella was also built out and demoed at ETHIstanbul, taking the WhatsApp trading-psychology agent in front of a hackathon audience outside its usual production surfaces.