Member Assistant: An Open-Source AI Helper for Mall Mini-Programs

Description and use

An AI assistant for mall mini‑programs that delivers instant, cited answers to member queries while keeping data on‑prem and enforcing read‑only safety constraints.

Problem

Mall mini-programs need instant, privacy‑preserving answers to member questions about points, parking, coupons, and orders without exposing sensitive data or incurring high costs.

Outcome

The assistant provides accurate, cited answers in milliseconds, keeps all member data on‑prem, eliminates per‑token costs, and enforces read‑only safety constraints in code.

Workflow

• 1. Member sends a question with X‑Member‑Id header to /chat
• 2. Service authenticates via header, retrieves relevant chunks from FAISS index
• 3. Optional preset tool call fetches data from read‑only SQLite
• 4. Assistant streams answer with citations over SSE
• 5. Low‑confidence answers fall back to a human agent

Evidence

Member Assistant is a production‑grade AI assistant for shopping‑mall mini‑programs. Members ask questions in natural language — "How many points do I have?", "How does parking work?" — and get instant, cited answers over an SSE stream.

Source

Real vector RAG — multilingual sentence‑transformers embeddings + FAISS index over the mall's knowledge base (parking rules, points rules, coupons, floor guide)

Source

Read‑only member tools — points / coupons / orders, served from SQLite opened in read‑only mode; the assistant can only call a closed registry of preset tools, never raw SQL

Source

FastAPI + Docker — /healthz, JSON structured logging, env‑based config, one‑command deploy

Source