AI-assisted Customer Support System

Tasks

• •Consolidated draft replies and tag suggestions from PHP and n8n into Spring AI, with human support staff reviewing and sending replies
• •Fetched customer context in parallel in the backend and separated fast draft generation from FAQ and document-based QA to reduce waiting time
• •Used Pub/Sub events, per-ticket Redis locks, and customer-input identifiers to prevent duplicate drafts and stale-event processing while generating new drafts for follow-up inquiries
• •Added error diagnosis using logs, traces, and user activity, together with citations and confidence evaluation, so support staff can review evidence and unresolved questions

Achievements

• •Observed an approximately 43% reduction in CS handling volume in operational data after system adoption

Related Posts