This demo simulates a customer support AI review workflow. It is not connected to a real service desk; instead it mocks the ticketing API, customer profile lookup, ticket history lookup, SOP knowledge base, agent draft generation, human review actions, and audit logs.

Tech used: Next.js, React, TypeScript, Next.js API Routes, DeepSeek API, LangChain JS for SOP text splitting and retrieval flow, bilingual default SOPs, uploadable SOP files, structured logging, PM2, nginx, and HTTPS deployment.

How to try it: generate a sample ticket or create your own ticket, run the agent, review the suggested response, inspect the matched SOP chunks and review checks, then approve, edit, escalate, or discard the draft.

The app does not expose hidden model chain-of-thought. Instead, it shows auditable operating evidence: the retrieved SOP chunks, category, sentiment score, route, confidence, decision summary, and processing log.

The interface, sample tickets, and default SOPs support English and Chinese. Uploaded SOPs can be marked as English, Chinese, or auto-detected so the demo can retrieve same-language and fallback policy matches.