AI Customer Support Agent with Tool-Calling & Mock CRM

Built a FastAPI-based customer support agent that uses LLM tool-calling to look up orders, retrieve customer info, and escalate or process refunds against a mock CRM (SQLite). Implemented a full agent loop: the model decides when a tool is needed, Python executes it, and the result feeds back into the conversation. Added session-based memory with automatic summarization for long conversations, plus guardrails to keep the agent on-topic. Designed as the foundation for tool-calling patterns reused in later multi-agent work.