This project delivers an AI-assisted, Germany-focused tax guidance experience in natural language. Purpose: help residents and workers understand deductions, rules, and tradeoffs without wading through static FAQs alone. Problem it solves: tax information is fragmented, jargon-heavy, and highly conditional on personal facts—generic search answers are often wrong or incomplete. How it works: a Next.js frontend talks to a FastAPI backend that runs a two-phase LangGraph pipeline (safety and routing, then extraction, optional clarification, optional Chroma RAG, and OpenAI generation), persists structured conversation state in SQLite, and streams answers over /chat/stream. Impact: users get grounded, conversational explanations with continuity, usage transparency (tokens/cost estimates), and clear informational boundaries—not a replacement for a tax advisor or official filings.
