Back to projects
AI AgentsRAGEditorial Tools

Plain Language Editorial Suite

A source-grounded editorial system that turns technical medical articles into accessible summaries for non-specialist readers.

Role
AI workflow and backend engineering
Focus
AI Agents / RAG
Stack
Python, LLMs, RAG

Context

The shape of the problem

Plain-language summaries are valuable because they make technical medical content understandable outside a specialist audience, but they are expensive and slow to produce manually. The constraint was quality: summaries needed to stay faithful to the source and fit an approval process involving authors and publisher-side reviewers.

Build

What I actually built

I built an AI-assisted editorial workflow around grounded generation, review states, and factual checks. The system drafts accessible summaries from article content, keeps the output tied to source evidence, and supports human approval rather than bypassing it. The product value is in accelerating the expensive first draft while keeping editorial accountability intact.

Engineering

Key decisions

01

Making complex medical writing accessible without drifting from the source article

Built a source-grounded generation flow that keeps outputs tied to article evidence

02

Supporting author and publisher approval rather than producing unchecked summaries

Added review and approval states so editorial users can inspect, amend, and accept generated drafts

03

Designing the workflow around editorial speed while preserving factual grounding

Structured the prompt and evaluation logic around clarity, factuality, and domain-appropriate simplification

Outcome

What changed

  1. 01Substantially reduced the manual effort needed to create plain-language summary drafts
  2. 02Kept the publisher and author approval process central to the workflow
  3. 03Created a margin-improving editorial automation pattern for a high-value publishing service

Stack

Tools and systems

PythonLLMsRAGDocument ProcessingEditorial ReviewEvaluation

I build delightful systems.

© 2026 Kristian Ernst.