/projects$cat support-automation.html

Support Automation

iLabs · 2025 · CloudofGoods customer ticketing

• Google ADK
• RAG
• Trino
• Milvus

the problem

Support agents spend most of their time writing the same kinds of replies — but each one needs this customer's order, this situation, and the company's actual service guidelines. Generic templates don't cut it; full manual drafting is slow.

what i built

An AI layer for the ticketing platform, built on Google ADK, that generates context-aware draft responses in seconds. It assembles context from two sources:

• Live data — customer and order details queried in real time through a Trino lake query engine.
• Memory — high-accuracy semantic RAG over a Milvus vector store to pull historical conversations and matching service guidelines.

The agent drafts; a human approves. Faster responses, consistent tone, grounded in real data.

It reuses the same Trino + Milvus retrieval layer as the Insight Agent.