interview-ingest

# interview-ingest

Audio transcription and document conversion for qualitative data import. Converts interview recordings, PDFs, and other formats into analyzable markdown.

## When to Use

Use this skill when:
- User has audio interview recordings to transcribe
- User needs to convert PDF documents
- User mentions "import data", "transcribe", "convert"
- Starting data preparation for Stage 1 or Stage 2

## MCP Dependencies

This skill operates at two capability tiers:

### Tier 1: Best (Requires MinerU API key)
- **PDFs:** MinerU VLM-powered parsing (90%+ accuracy)
- **Tables/Images:** Excellent extraction
- **Audio:** External transcription services recommended
- **Best for:** Complex academic papers, documents with tables/figures

### Tier 2: Manual (No API key required)
- **PDFs:** Manual conversion (Adobe Acrobat, Google Docs OCR)
- **Audio:** External transcription services (Otter.ai, Rev.com, YouTube captions)
- **Tables/Images:** Manual cleanup after conversion
- **Best for:** Simple documents, researchers without API keys

## Checking Tier Availability

```bash
# Check for MinerU
[ -n "$MINERU_API_KEY" ] && echo "MinerU available (Tier 1)" || echo "Manual conversion (Tier 2)"
```

## Workflow by Format

### Audio Interviews

**Recommended transcription services:**
- **Otter.ai** - AI-powered transcription, good for interviews
- **Rev.com** - Professional human transcription
- **YouTube** - Upload as unlisted video for auto-captions
- **Whisper** - Open source, run locally

**Best practices:**
- Use high-quality recordings when possible
- Review transcripts for accuracy
- Add speaker labels: "Interviewer:" and "Participant:"
- Note timestamps for key passages
- Mark unclear passages with [unclear] or [inaudible]

### PDF Documents

**Tier 1 (MinerU - recommended for complex PDFs):**
```bash
# Parse PDF with VLM mode for tables/images
mineru_parse({
  url: "file:///path/to/paper.pdf",
  model: "vlm",
  formula: true,
  table: true
})
```

**Tier 2 (Manual co