analyze-interviews

Analyze Interview Transcripts

Analyze all interview transcripts in a directory, extract themes, rank insights, and generate a comprehensive report.

Step 1: Discover and Validate Transcripts

Scan the transcripts directory ($2):

• Find all text files (.txt, .md, .doc, .docx)
• Count total transcripts
• Calculate total file size and estimated token count
• Identify any context files (ICP, research goals, methodology docs)

Look for context file ($3):

• If --context-file argument provided, read that file
• If not provided, look for common files in directory:
  • icp.md or ideal-customer-profile.md
  • research-goals.md or objectives.md
  • methodology.md
  • customer-profile.md or persona.md

Report to user:

## 📁 Transcripts Discovered

**Directory:** $2
**Transcripts found:** [number] files
**Total size:** [size in MB]
**Estimated tokens:** ~[number]k tokens

**Files:**
- [filename 1] ([size])
- [filename 2] ([size])
- [etc.]

**Context documents found:**
- [ICP/research goals/methodology if found, or "None"]

Would you like to proceed with analysis?

Wait for user confirmation.

---

Step 2: Determine Agent Scaling Strategy

Based on workload, intelligently determine how many parallel agents to use:

Agent Scaling Logic:

If 1-3 transcripts:
  → Use 1 agent (sequential analysis is fine)

If 4-8 transcripts:
  → Use 3 parallel agents
  → Divide transcripts evenly (e.g., Agent 1: transcripts 1-3, Agent 2: 4-6, Agent 3: 7-8)

If 9-15 transcripts:
  → Use 5 parallel agents
  → Divide transcripts evenly

If 16+ transcripts:
  → Use 8 parallel agents (optimal for most systems)
  → Divide transcripts evenly

ALSO consider total token count:
If total tokens > 150k:
  → Increase agent count by 1-2
  → Example: 6 transcripts but 200k tokens → use 5 agents instead of 3

If total tokens < 30k:
  → Decrease agent count
  → Example: 10 short transcripts with 25k tokens → use 3 agents instead of 5

Present strategy to user:

## 🤖 Agent Strategy

Based on [number] transcripts totaling ~[number]k tokens, I recommend using **[number] parallel agents** for efficient analysis.

**How it works:**
- Each agent will analyze [number] transcripts
- Agents work simultaneously (faster than sequential)
- All findings will be synthesized into final report

**Agent assignments:**
- Agent 1: [transcript names]
- Agent 2: [transcript names]
- [etc.]

**Estimated time:** [minutes] minutes

Would you like to:
1. ✅ Proceed with this strategy
2. 🔄 Adjust agent count (tell me your preference)
3. ❌ Cancel