Khora AI

Khora AI

Ask about the training, the timber assemblies or the scores. Describe something you would like built and Khora will read the request back to you, ask what it needs to know, and then run the chain.

Training Process

V

Visitor's Khora

Sign in to keep your own record here.

The more you chat or assign tasks, the closer you grow

Khora can learn and grow on its own

With this on, this browser keeps a record of the days you visit and the questions you ask, and the numbers above start moving. It is stored on this device only — nothing is sent anywhere.

Inspiration

The briefs this system was actually given. Tap one and the chain that came out of it plays back in full — what it understood, the questions it asked and the answers it was given, the brief it expanded them into, then the assembly and the review. It is a recording of a run that really happened, so watching it costs nothing.

At the foot of a replay, Add to new chat puts that design's own expanded brief into the message box. It is not sent — edit it first, and send it when you mean to.

Khora Advantage

The same brief, given to Khora and to a general-purpose model. Four generations of Khora are on record here, each set against a different counterpart — Gemini, Deepseek, Claude Code, KIMI.

What the spreads measure is not taste. It is whether the thing came back buildable: every member drawn from timber that actually exists in the stock, no member reused, nothing left floating. Where a counterpart failed, the error it failed with is printed underneath it. The token figures are what each run cost.

About

Khora AI is a design assistant built for Palimpsest — RC5, The Bartlett School of Architecture, UCL.

Palimpsest asks what can be built from waste material that actually exists. It comprises three products: an interactive web page, an LLM system, and 3D-printed connectors. The wider project is at palimpsest-2026.vercel.app.

Khora AI is the design assistant built on the LLM system.

Khora AI

Main designed and developed by Jiang Linyun

  • Every training round, and the scoring routine that reviews each design it produces
  • The chain architecture and its server
  • The interface, the interaction design, and the front-end you are reading this in
  • The exhibition site this page belongs to

Palimpsest — the team

Acknowledgements

The idea of an assistant trained on the studio's own timber stock was first raised in discussion with Marios Tsiliakos, Stefan Bassing and the team, May 2026. Compute for the training rounds was covered by Jiang Linyun; the token costs of using it during the Palimpsest project were shared by the team.

Select project Design assistant ready