Blog

Notes on support, AIand the people in between.

How we build Kuvra, what we got wrong on the way, and the parts of running support that no tool fixes for you.

14 July 2026 · 9 min read

Building a RAG pipeline that actually answers the question

Dense vectors alone miss the exact phrasing customers use. Here's how we combined semantic search, keyword search and reciprocal rank fusion — then added a reranker — to get support answers grounded in the right paragraph.

Read article

All articles

4 articles

Product

2 July 2026 · 6 min read

The handoff problem: moving a chat from AI to human without losing context

Most AI support tools get the answer right and the handoff wrong. The moment a person is needed is the moment the experience usually falls apart — here's how we designed around it.

Read article

Playbook

21 June 2026 · 7 min read

Four support metrics worth watching (and three that mislead you)

Most support dashboards optimise for the number that's easiest to compute. Here are the four we'd actually run a team on, and the three that quietly reward the wrong behaviour.

Read article

Security

9 June 2026 · 11 min read

Hardening a public chat widget: XSS, origin allowlists and prompt injection

A support widget sits on the open internet, accepts untrusted input, renders it in an authenticated dashboard, and forwards it to a model with access to your docs. Four surfaces, four fixes.

Read article

Playbook

30 May 2026 · 5 min read

Writing help docs an AI can actually retrieve from

Retrieval doesn't read your documentation the way a person does — it fetches passages. A few small habits make the difference between an agent that answers and one that escalates everything.

Read article

Notes on support, AIand the people in between.

Building a RAG pipeline that actually answers the question

All articles

The handoff problem: moving a chat from AI to human without losing context

Four support metrics worth watching (and three that mislead you)

Hardening a public chat widget: XSS, origin allowlists and prompt injection

Writing help docs an AI can actually retrieve from

Something you'd like us to write about?A person reads every message.