lorraine

Purpose: Replace the existing research search UI with a simple, guided Q&A flow that displays many source wordings (formal, business, casual, social, etc.) and captures how the user would naturally say the same thing. Collected answers form a small dataset that can be used to fine‑tune a LoRA on the user's casual style.

Goals

• Present prompts from a JSON file (no model required to ask questions)
• Capture answers via text input and optional voice dictation (Groq Whisper)
• Keep the clean, minimal design from the current app (same fonts, spacing, Tailwind) and keep the API key UX
• Export answers as JSON for downstream fine‑tuning

Data Model

• Prompts JSON: array of items
  • id: string
  • label: short descriptor
  • instruction: what to do (e.g., "Rewrite this in your style")
  • source_text: the wording to adapt
  • tags: string[] (optional)
• Answers (client-side):
  • question_id: string
  • answer_text: string
  • transcript_text?: string (if captured via Whisper)
  • created_at: ISO timestamp

Frontend Flow

• Load prompts from /data/questions.json
• Show current prompt card (instruction + source text)
• Provide a text area for "Your version"
• Optional mic button to record and transcribe with Whisper; user can insert the transcript as their answer
• Navigation: Previous / Next; progress indicator
• Export: Download JSON of all answers
• Persist API key (Groq) in localStorage; server key supported if present

Backend Endpoints (Hono)

• GET /data/questions.json → serve local JSON file (Val Town compatible loader)
• POST /api/transcribe → forwards uploaded audio to Groq Whisper (model: whisper-large-v3) and returns { text }
• Existing /api/check-key retained for client key handling

Export Format

{
  "app": "lorraine",
  "version": 1,
  "answers": [
    {
      "question_id": "q1",
      "answer_text": "...",
      "transcript_text": "...",
      "created_at": "2025-01-01T00:00:00.000Z"
    }
  ]
}

Notes / Future

• Add prompt categories and filters
• Add session save/restore in localStorage
• Add CSV export and metadata like language/setting
• Add batch microphone capture (hands‑free), and per‑answer audio blob export if needed