Moda is an AI-native design platform for non-designers like marketers and entrepreneurs. The goal is to create professional-grade presentations, social media posts, and brochures.

Moda Architecture

At the core of Moda is a multi-agent system built with Deep Agents, with LangSmith providing the observability to iterate quickly. The system consists of three main agents:

  1. Design Agent: handles design creation and iteration.
  2. Research Agent: fetches and stores structured content from external sources.
  3. Brand Kit Agent: ingests brand assets (colors, fonts, logos) from websites or existing documents.

The Research Agent and Brand Kit Agent use Deep Agents. The Design Agent, initially based on a custom LangGraph loop, is being migrated.

Context Engineering: The Details That Matter

To achieve high-quality results, Moda has implemented a context engineering system that includes:

  • Custom DSL: Instead of using a raw scene graph, Moda developed a DSL (Domain Specific Language) that provides a cleaner, more compact representation of the canvas, reducing costs and improving output quality.
  • Triage → Skills → Main Loop: Each request passes through a triage node that classifies the output format and loads the relevant skills (Markdown documents with best practices and guidelines). Skills are injected as human messages, with prompt caching to optimize performance.
  • Dynamic Tool Loading: The design agent uses a set of core tools, with the ability to activate additional tools on demand.
  • Scaling Context to Canvas Size: Moda dynamically manages the amount of context provided to the agent, offering a complete overview for small canvases and a high-level summary for larger projects.

UX: Human-AI Collaboration

Moda's user interface is designed to foster collaboration between user and AI. Instead of a workflow based on generation and replacement, the AI works directly on a fully editable 2D vector canvas.

Observability with LangSmith

LangSmith provides full visibility into agent execution, facilitating prompt and tool iteration, cost tracking, cache analysis, and error diagnosis.

Moda is considering implementing formal evals, but for now, LangSmith traces are the primary feedback tool.