Inspiration...

Single-prompt AI generators often produce inconsistent, repetitive, and context-unaware content, lacking memory of previous outputs. We aimed to create a professional writing workflow that overcomes these limitations.

## What it does
"The Master Builder" employs a four-agent architecture:
1.  **Scribe:** Drafts content based on high-level goals.
2.  **Editor:** Refines the Scribe's output for quality and consistency.
3.  **Archivist:** Maintains and updates a persistent world state/lore document.
4.  **Inspector:** Ensures the output aligns with the world state and narrative progression, requesting revisions if needed.
This system collaboratively maintains world state and narrative progression across chapters.

## How we built it
We used React 19, TypeScript, and Vite for the front-end. The core logic utilizes Google's Gemini 3 Pro and Flash models, leveraging structured JSON outputs for inter-agent communication and specific thinking budgets to control depth and speed. LocalStorage is used for persisting the world state locally.

## Challenges we ran into
We encountered challenges with the AI service layer, including implementing exponential backoff strategies to handle rate limits effectively and reliably extracting JSON from the models' responses.

## Accomplishments that we're proud of
We are proud of the successful multi-agent orchestration loop and the system's ability to maintain a persistent world-state, ensuring narrative consistency over time.

## What we learned
This project highlighted the importance of deep reasoning capabilities and well-defined, structured schemas for building reliable and predictable AI agents that can collaborate on complex tasks.

## What's next for The Master Builder

"We plan to continue expanding the system's ability to handle the creation of full-length novels of 88+ chapters, further refining the agents' interactions and world-state management."

Built With

Share this project:

Updates