All work An e-commerce software company
An AI assistant that answers support tickets in three languages
Placeholder case study. An assistant that drafts answers from a company's own help centre, with a human approving every reply.
- Client
- An e-commerce software company
- Year
- 2026
- Services
- AI, Backend, Integrations
- Stack
- TypeScript, Postgres, pgvector
- 62%of tickets answered from a draft without editsReported by the client
- 3languages supported at launch
- <2 smedian time to a draft
This is placeholder content for the scaffold. The owner replaces it with a real, approved case study.
The problem
A growing support team answered the same questions every day, in English, Romanian and Russian, while the answers already existed in the help centre.
The approach
We built an evaluation set from a thousand past tickets before writing the assistant, so every change could be measured against real questions.
What we built
A drafting assistant inside the existing help desk: it retrieves the relevant help-centre articles, drafts a reply in the customer’s language and shows its sources. A person approves or edits every reply.
The result
Most tickets are now answered from a draft without edits, and new support staff reach full speed in days rather than weeks.
