An AI support assistant grounded in company data
Built a retrieval-augmented assistant that answers shipment and policy questions using live operational data.
- Industry
- Logistics
- Timeline
- 10 Weeks
- Platform
- Web & API
- Live website
- Private beta
41%
tickets auto-resolved
Routine shipment questions are answered without a human.
6 min
faster responses
Customers get grounded answers in seconds, not hours.
0
data leaving the VPC
Everything runs inside the client's own cloud account.
400
evals per release
Every release is checked against real customer questions.
Placeholder quote — our support team finally trusts the answers an AI gives.
The challenge
Cargoware's support team spent most of its day answering the same questions about shipment status, customs paperwork and policies, while customers waited hours for replies.
Our approach
We built an assistant that combines semantic search over policy documents with tool calls into the live shipment API. Every answer cites its sources, and anything uncertain is escalated to a human with full context.
- Evaluation suite of 400 real questions run on every change
- Role-based access so the assistant only sees data the user may see
- Human-in-the-loop review dashboard for the support team
The outcome
Within two months the assistant was resolving a large share of incoming tickets on its own, and the support team could focus on complex cases.