Skip to content
Silka

AI & research2026EN

NOVA

Seven surfaces in one workspace and 149 citations across 24 sources — no answer you cannot open and check.

Visit the live site
  • SaaS
  • Product UI
  • AI
  • Design system
NOVA home screen with chats, documents and the credit meter
Industry
AI & research
Engagement
Ongoing
Team
1 designer · 1 engineer
Platforms
Web app · Chat · Document library · Workspace search

01

Challenge

An assistant that answers confidently without showing where the answer came from is useless for research. You cannot quote it in a report, you cannot check it, and six months later you will not remember whether it was in a document or the model decided it.

NOVA had to be a place of work rather than a chat with files attached: documents go in, questions get asked, and what comes out stays — as a saved insight, as a project with open questions, as a source carrying a count of how often it has been leaned on.

Corpus
24 documents · 421 pages · 37 MB
Surfaces
7
Hard part
The answer has to be checkable, not persuasive

02

Solution

The citation became the unit of the interface. 149 references across 24 sources get a surface of their own: every source shows how many times it has been cited and in which conversations. “What is this based on” is a screen, not a search through chat history.

The chat shows what it actually read: “4 in context” sits beside the answer — not a metaphor, the number of documents that went in. The model budget is in plain sight too, in the sidebar: 18,420 of 25,000 credits. A product that spends money per call should not hide the meter.

03

Architecture

  1. L0
    WorkspaceClient
  2. L1
    ChatService
    LibraryService
    SearchService
  3. L2
    Context assemblyService
    Document indexData
  4. L3
    CitationsData
    ModelExternal
Connections (10)
  • Workspace Chat
  • Workspace Library
  • Workspace Search
  • Chat Context assembly
  • Library Document index
  • Search Document index
  • Context assembly Document index
  • Context assembly Model
  • Model Citations
  • Citations Chat
Everything the chat shows comes out of the same index the library fills — and comes back with a citation.

04

UX

Research does not fit in one list, so the workspace is split into seven surfaces: a home for what has been happening, chat, library, insights, search, projects and sources. Each answers a different question — what do I already know, what did I ask, what is it standing on.

A project here is a set of open questions rather than a folder: “2 of 4 research questions answered” sits on the card. That framing is what stops research turning into an archive of conversations with nothing finished.

Search
Ctrl K — documents, chats, insights, projects
New chat
Ctrl N
Projects
8, four of them active

05

UI

A dark surface and one warm accent. Orange marks exactly what starts work — “New chat” and the active section; everything else lives on greys, so an hour inside it does not tire the eye.

Density is set for reading: document and source rows carry format, size, page count and date on one line, with none of the air a card grid would demand. This is somewhere a shift gets worked, not a showroom.

06

Build

All seven surfaces are assembled from one component library: the list row, the metadata card, the counter, the tag filter and the detail panel repeat everywhere. Adding an eighth surface is assembly, not another screen drawn from scratch.

The states usually left until last are in from the start: an empty library, pinned conversations, the archive, a document still processing, and the credit budget run out.

07

What it does

  • Sources with counts

    24 sources, 149 citations: each one shows how often it is cited and where.

  • Context in the open

    The number of documents that actually went into context sits beside the answer.

  • Projects as questions

    Research runs as a list of questions with answered/open state, not a folder of files.

  • Search across everything

    Ctrl K searches documents, conversations, insights and projects at once.

  • Saved insights

    31 findings with tags — what is left once a conversation is closed.

  • A visible budget

    Model credit usage is in the sidebar, not buried in billing.

08

Integrations & stack

Integrations

  • PDF
  • DOCX
  • CSV
  • Markdown
  • Web pages

Frontend

  • Next.js
  • React
  • TypeScript
  • Tailwind CSS

AI

  • Retrieval
  • Citations
  • Context assembly

Product

  • Design system
  • Keyboard shortcuts
  • Workspace search

09

Where it landed

7
Workspace surfaces

Home · chat · documents · insights · search · projects · sources

149
Citations across 24 sources

Every one opens the page it came from

421 pages
Indexed

24 files, 37 MB

In plain sight
Model budget

18,420 / 25,000 credits in the sidebar

An AI workspace you can argue with: any statement is one click from the page that produced it — which is the whole difference between a research tool and a persuasive chat.

NOVA conversation list with pinned chats
Conversations
NOVA sources with citation counts
Sources and citations
NOVA research projects with progress
Projects
NOVA on mobile
Mobile

Every build is designed for the phone as its own experience, not as a narrowed desktop layout.

Got a product that should exist by now?Tell us about it.

Describe what has to change in your business. We will work out what that actually takes — including when it takes less than you think.