AI Historical Figures for Museums: Exhibits That Talk Back

A museum label can tell a visitor what a ship's logbook says, but it can't answer the eleven-year-old who wants to know what the crew ate when the biscuits ran out. With Neocortex an exhibit can answer that question in character: an AI historical figure or expert guide whose answers come from the museum's own vetted sources.

What it looks like

The Harbour Light Maritime Museum is an invented museum with a large archive of nineteenth-century merchant logbooks. Its new gallery has a kiosk shaped like a ship's binnacle, and on its screen stands Captain Ines Varga, master of the barque Aurelie. A small panel beside the kiosk explains that Captain Varga is a composite character voiced by AI, built from the museum's logbooks and letters.

A school group gathers round. One girl leans into the microphone:

Visitor: What did you eat on the long crossing? Captain Varga: Salt beef, hard biscuit and dried peas, mostly. By the sixth week the biscuit had weevils, so the cook knocked each one on the table before serving. Visitor: Did anyone fall overboard? Captain Varga: Our log records one seaman lost in a squall off the Azores. I won't tell you more than the log does.

As she mentions the log, the reply carries a SHOW_ARTEFACT action that brings up the scanned page on the kiosk's second screen. When a boy asks about something the archive doesn't cover, she says so and points him to the curator's desk instead of inventing an answer.

After a few minutes the kiosk asks if the group wants to test their sea legs. Three quiz questions later, the screen shows their score and the captain congratulates the winning table with a line in her weathered voice.

How Neocortex makes it work

• AI characters. Captain Varga is designed in the Neocortex visual node editor: personality, backstory, the voyage she commanded, how she speaks, and the actions she can return. Curators can question her in the browser dashboard long before the kiosk is installed.
• Knowledge base. Transcribed logbook pages, letters and the museum's own exhibit texts go in as PDF or TXT documents, so answers stay grounded in material the museum has checked. Free includes 1 upload, Starter 5, Pro 20 and Team 100.
• Custom voice design. Describe the voice in text ("a woman in her fifties, low and weathered, a slight northern accent") and get a voice to match. Pro includes 5 custom voices, Team 20.
• Emotions. Each line carries an emotion tag, so the captain looks grim when she talks about the squall and pleased when a child guesses the cargo.
• Actions. SHOW_ARTEFACT and PLAY_MAP come back with replies, and your kiosk code brings up the scan or animates the route.
• Conversation Flows. An optional guided mode walks visitors through the voyage with Start, Checkpoint and End nodes: departure, the storm, landfall. The returned flowState tells the kiosk when to move to the quiz. Flows are on every plan.
• Quiz engine. Question sets and scoring through the API end the visit with a short quiz (Team plan).
• More than 20 languages. Tourists can ask in their own language and the captain answers in it.
• Session memory. Keep the sessionId while a group stands at the kiosk, and start a fresh one when the next group arrives.

Ideas to build

The Roman shopkeeper. A classroom app for history lessons puts pupils in an invented shop on a Roman street, run by a cloth dyer called Septima. Pupils ask about prices, gods, the noise from the bakery next door. The knowledge base holds the teacher's own lesson notes, and a Conversation Flow makes sure every group asks about at least three topics before the quiz.

The science centre robot. Not every guide needs to be historical. An invented robot guide called Bolt explains the exhibits in a science centre, gets visibly excited about magnets and returns a START_DEMO action that sets the hall's electricity display running.

The web exhibit. A museum's online collection gets an expert guide who sits beside the gallery viewer. Visitors at home ask about a painting, and the guide answers from the catalogue entries while a ZOOM_TO_DETAIL action brings the brushwork into view.

The VR deck. On Meta Quest, built in Unity with the XR Interaction Toolkit, visitors stand on the deck of the Aurelie in a storm and ask the captain what to do. Lip sync from the Unity SDK moves her face as she shouts orders over the wind.

Building it

Kiosks run well as Unity desktop apps through the Unity SDK, web exhibits use Unity WebGL or call the Web API from any stack, and VR versions run on Meta Quest through Unity. For voice in the browser, use the Unity WebGL microphone (over HTTPS) or your own mic capture with Neocortex speech-to-text.

1. Choose the character carefully. An invented or composite figure, or a clearly labelled portrayal of a real one, with a panel saying how it was made.
2. Assemble and review the sources. Curators pick the documents the character may draw on and check them before upload.
3. Create the character in the visual editor with a personality, the knowledge base, a designed voice and a few artefact actions.
4. Test hard in the dashboard. Ask the questions children ask, the questions sceptics ask, and the questions the sources can't answer.
5. Connect the kiosk. Microphone in, voice out, actions mapped to screens and media, a new sessionId for each group.
6. Add the quiz on the Team plan, or close with your own activity screen.

We recommend that the institution owns the source material and reviews it on the same schedule as any other gallery text. If a document would not pass as a wall label, it should not be in the knowledge base.

Museum characters are one entry in our AI NPC, ambient AI and voice command use cases. For other learning ideas read AI conversation partners for language practice and VR soft skills training with AI. The education and training page and the knowledge bases feature page cover the rest. Start with one character, one voyage and the documents you trust most.

Checked October 2026.

Frequently Asked Questions