Hyve Chatbot

Role and ownership

I led design on this account, from definition through active development.

Decisions, and what we set aside

A focused entry point rather than an open chat window. A conversation workspace that keeps retrieved documents beside the answer. One dark interface across dashboard, chat and sources. A product language that stays coherent when the client identity changes.

The interface did not arrive in one pass: two earlier versions were designed and set aside before the one that shipped.

What it had to hold: private, client-specific deployments; company knowledge as the only source; source material visible alongside the answer; assistants, prompts, conversations and knowledge resources in one workspace; an identity system that adapts per client.

What was delivered

Core product screens

84 artboards

Connected surfaces

Dashboard · chat · sources

Counts describe the scope of the work, not its results.

Where it stands

The initiative was later discontinued as an internal product.

What I learned

Designing a knowledge assistant is where I learned what an LLM product really is. The model is the easy part to demonstrate; the design work sits in provenance: what the interface can honestly claim, and how a person checks an answer instead of taking it on faith.

The work 2 views

1. 01
   Dashboard

   The assistant first, then the question

   The argument was whether to open with the question or with the assistant, and the assistant won. Then what was asked before, what it can draw on, and the prompt itself: a starting point rather than a menu.

2. 02
   Conversation

   The answer and its sources, in one view

   Putting the source beside the answer was a requirement, not a preference. An answer is worth little if checking it means leaving the chat, so the source sits open and downloadable where the doubt actually arrives.