PromptEvolver v10 - Gene Splicer

Shopping Query

Seed CQs (5 required, one per line)

0 / 5 CQs

Gemini API Key (optional with proxy)

Use CORS Proxy

Evaluation Mode

LLM-as-Judge AI rates everything automatically Human Evaluator You rate everything manually Human → LLM Handoff You rate first, then hand off to AI

Generations 5 recommended for human

📜 Warm Start (optional)

Paste prescriptive rules to guide evaluation. Use DO/DON'T/KILL language. Leave empty for cold start.

Strategy Guidance (what to ask)

Style Guidance (how to phrase)

Generation

1 / 5

Mode

Human

Query

Kept

Killed

R:Bad = Kill lineage

R:Mid or S:Bad/Mid = LLM fixes

R+S:Good = Keep (mutates)

Preserve = Locked (done)

Select text to annotate snippets →

🔍 Pattern Signals 0

Select bad phrases from CQs...

📊 Session Rationales (Gemma 3 27B)

Strategy

(learning...)

Style

(learning...)