Skip to content

kickoff.md

Latest commit

 

History

History
131 lines (112 loc) · 3.63 KB

kickoff.md

File metadata and controls

131 lines (112 loc) · 3.63 KB

pauseai-l10n

This project provides LLM-based localization (l10n) support for the pauseai-website project, maintaining a version-controlled cache of translations and the prompts that generated them.

(This doc is a summary by Claude. this note is added by Anthony: I don't expect the suggested structure and naming conventions to persist. Consider the structure as an illustrative example.)

Architecture

The project serves as a:

  1. Cache for LLM-generated translations
  2. Version control for translation prompts
  3. Build-time integration for static site generation

Translation Cache Structure

pauseai-l10n/
├── cache/
│   ├── en-US/                    # Source content
│   │   └── [content-hash].[human-clues].json   # Original content
│   ├── es-ES/                       # Spanish translations
│   │   └── [content-hash].[human-clues].json   # Contains translations and metadata
│   └── ...                       # Other languages
├── prompts/
│   ├── default.txt               # Base prompt template
│   └── locale-variants/          # Optional locale-specific additions
│       ├── es.txt
│       └── de.txt
├── src/
│   ├── hash.ts                   # Content hashing utilities
│   ├── prompt.ts                 # Prompt generation/management
│   ├── translate.ts              # Model-agnostic interface
│   ├── models.ts                 # Simple model selection/configuration
│   └── cache.ts                  # Cache management
└── tests/

Translation Flow

  1. During website build:
    • Hash each source content piece
    • Check cache for existing translation
    • If missing/outdated, generate new translation via LLM
    • Store result in cache
    • Generate static pages for each locale

Cache Entry Format

{
  "sourceHash": "hash-of-original-content",
  "sourceContent": "Original en-US content",
  "translations": [
    {
      "content": "Translated content",
      "timestamp": "2024-03-20T12:00:00Z",
      "model": "gpt-4",  // Simple identifier for tracking/analysis
      "prompt": {
        "template": "default",
        "version": "v1.2"
      },
      "feedback": [
        {
          "source": "discord",
          "comment": "Good translation but formal tone",
          "timestamp": "2024-03-21T14:30:00Z"
        }
      ],
      "status": "active"
    }
  ],
  "metadata": {
    "path": "about/mission.md",
    "type": "page",
    "lastModified": "2024-03-20T12:00:00Z"
  }
}

Development Priorities

  1. Core Infrastructure

    • Content hashing mechanism
    • Cache storage/retrieval
    • Model-agnostic LLM interface
    • Model-specific adapters

  2. Prompt Engineering

    • Base prompt template
    • Page-specific additions
    • Model-specific additions
    • Locale-specific additions
    • Some additions for particular combinations of those
    • Whole-page translation focus

  3. Build Integration

    • Svelte/Netlify integration
    • Incremental translation
    • Preview system

  4. Monitoring/Validation

    • Translation quality metrics
    • Cost tracking
    • Cache statistics

Getting Started

  1. Set up development environment:

    npm init
npm install @types/node openai

  2. Create basic cache structure:

    mkdir -p cache/{en-US,es,fr} prompts src tests

  3. Link for local development:

    npm link
cd ../pauseai-website
npm link pauseai-l10n

Next Steps

  1. Implement content hashing
  2. Create basic cache management
  3. Develop initial prompt template
  4. Set up LLM integration
  5. Create build-time hooks
  6. Add monitoring/logging