Skip to main content

​
Problem

First drafts are rarely perfect. You want the LLM to generate content, critically evaluate it, and produce an improved version based on feedback.

​
Solution

The self-refinement loop uses a three-stage process:
  1. Generate: Create an initial draft
  2. Critique: Analyze weaknesses and areas for improvement
  3. Improve: Produce a refined version incorporating the feedback
Each stage is a separate job, allowing the model to focus on one task at a time.

​
Example

Refine a cover letter: 
GOAL="Write a cover letter for a senior engineer position"

# First draft
draft=$(wrk ./workspace "$GOAL" | xargs flw ./workspace)

# Critique
critique=$(wrk ./workspace "Critique this draft. What's weak? $draft" | xargs flw ./workspace)

# Improve
final=$(wrk ./workspace "Improve this draft based on feedback:
Draft: $draft
Feedback: $critique" | xargs flw ./workspace)

echo "$final"

​
How It Works

  • wrk ... | xargs flw submits a job and immediately waits for the result
  • Each stage uses the previous output as context
  • The critique stage provides specific, actionable feedback
  • The improvement stage has both the original and the critique

​
Variations

Multiple iterations: 
for i in {1..3}; do
  critique=$(wrk ./workspace "Critique: $draft" | xargs flw ./workspace)
  draft=$(wrk ./workspace "Improve based on: $critique\n$draft" | xargs flw ./workspace)
done
Specialized critique: 
# Critique from different perspectives
clarity=$(wrk ./workspace "Rate clarity: $draft" | xargs flw ./workspace)
accuracy=$(wrk ./workspace "Rate technical accuracy: $draft" | xargs flw ./workspace)
style=$(wrk ./workspace "Rate writing style: $draft" | xargs flw ./workspace)

# Improve with all feedback
final=$(wrk ./workspace "Improve this:\nDraft: $draft\nClarity: $clarity\nAccuracy: $accuracy\nStyle: $style" | xargs flw ./workspace)

​
Use Cases

  • Writing: Essays, documentation, cover letters
  • Code review: Generate code, critique, refactor
  • Design: Create designs, evaluate, iterate
  • Strategy: Draft plans, identify gaps, revise

Build docs developers (and LLMs) love

Get started for freeTalk to us