E-Commerce AI Customer Service Agent (FastAPI + Dify)

E-Commerce AI Customer Service Agent is a production-ready, plug-and-play

support agent for online stores. It runs the full pipeline — input redaction,

intent recognition, persona matching, knowledge retrieval, reply generation,

compliance guard, and human handoff — and ships in two interchangeable forms:

a pure-Python FastAPI backend and an importable Dify low-code workflow

(18 nodes / 19 edges). Both paths share one source of truth (YAML policies +

Markdown prompts), so your operations team can tune personas, intents,

compliance rules and handoff thresholds without touching a line of code.

Highlights:

• 7 customer personas (VIP, price-sensitive new buyer, anxious order-chaser,

angry complainant, researcher, silent recall, fallback) matched by weighted

tag rules.

• 8 intents recognized via keyword scoring with an LLM fallback on low

confidence.

• BM25 knowledge retrieval (top-3 recall) — fully offline, zero database.
• Built-in compliance guard (advertising-law wording, over-promised

compensation, competitor defamation, privacy, medical claims) with

block / rewrite-retry / local sanitize — risky copy never ships.

• Automatic human handoff with a 5-element ticket summary (4 trigger types).
• Privacy-first: phone & ID numbers are masked before any model call.
• Offline fallback: runs without an LLM API key using template generation;

the pipeline never breaks.

• 23 unit tests, one-command demo, interactive Swagger docs at /docs.

Ideal for e-commerce stores, SaaS support teams, and developers who want a

safe, explainable, self-hosted AI support agent they can customize in YAML.