transaction-classification

Hybrid GenAI Transaction Categorization is a privacy-first AI system that automatically classifies natural language financial transactions into categories and merchants. It uses a hybrid pipeline combining an ONNX-optimized DistilBERT classifier with an LLM fallback for low-confidence cases, ensuring both speed and reasoning capability. The system

Two-level transaction categorisation pipeline for a Spanish fintech aggregator. LightGBM L1 (91.2% accuracy, 16.8ms) escalates to Azure OpenAI L2 only when confidence drops. GDPR-safe via Presidio + regex anonymisation. FastAPI + Streamlit.

Privacy-first transaction categorization system using rules, ML, and LLM ensemble routing.

A free, open-source, local-first personal finance workbench for power users.

Current-state Bitcoin mempool viewer. Polls one to four nodes, classifies transactions through independent lenses (exact properties, shape heuristics, data protocols, BIP-110 policy), and compares node mempools in the browser.