Alcade

I wanted to play D&D with AI. So I built it, then thought: might as well pitch it.

Timeline: Late 2022 - mid 2023
Mine: The prototypes, the game design, the deck
Studio: An Etherate Labs in-house bet; Ishaan and Junaid joined the pitch
Result: One playable chapter, three playtesters, one grant deck

It was the end of 2022, the tech was weeks old, and nobody could tell me it wouldn’t work because nobody knew anything yet. So I started building small prototypes to find the game.

Six AI-generated panoramas of the same wizarding market street in different styles - daylight painting, neon night, 3D animation, dark storefronts, castle town, photoreal - each labeled with the raw prompt filename it was generated from — The same market street, six ways. The filenames are the prompts; finding the game started with finding what the models could draw.

The build

One playable chapter. You describe a character, the AI writes the backstory, paints the portrait, and drops you into a town with a mystery.

The game was the excuse; the remembering was the project. An NPC with its own objectives and backstory holds a conversation, keeps notes on you, and lays a breadcrumb trail toward the mystery. I wired up character creation, scene art generated on the fly, and a memory loop so the town didn’t forget you between questions.

Flow diagram of the Alcade memory loop: user input analyzed by an LLM into embeddings and entities, compared against a search index, retrieved documents assembled into a prompt, and the search index and entity store updated after every generated reply — The part I cared about: characters that remember you. What an NPC knows, what it learns from you, what it stores.

The pipeline, live: three words in, a life story and a portrait prompt out.

The playtest

Three of my friends playtested it. They promptly broke out of the character system and got the NPCs doing whatever they wanted. There wasn’t a word for that yet. My friends found it in minutes anyway.

Slide from the Alcade pitch deck showing the character creation form, with a pink Streamlit error bar at the bottom reading “NameError: name ‘response’ is not defined” — Spot it: the pitch deck’s own screenshot went out with a live NameError.

The pitch

Then I thought, might as well pitch it. We wrapped it in an Etherate deck and took it to a couple of blockchain networks, Mantle among them. Nothing came of it, and Etherate wound down that August. One chapter, three playtesters, and the prototypes did what I built them for: I got to play D&D with AI.