A retired chemist turned a newsletter archive into an AI that writes book
A retired analytical chemist built an AI pipeline that turns French herbal newsletters into a research database and a full, evidence-ranked book chapter.
A retired analytical chemist sat on a personal collection of French-language phytotherapy newsletters and a dream of a book. In the program they built an AI agent that reads each letter, files the science into a relational database and writes chapters to their own strict standards. It is a story of a scientist's rigour meeting new tools, and of choosing real sources over easy volume.
Before
They had a personal collection of phytotherapy newsletters in French, along with a plan for a book on herbal health. The material was rich but scattered across PDFs and languages. Turning it into something structured and trustworthy, with a chemist's standard for evidence, was a large manual job.
“This project turns that collection into a self-sustaining pipeline: source newsletter → structured extraction → relational database → publication-quality book chapter, with an AI agent doing the work at every step.”
In their own words
The turning point
Lesson 9 gave them the Memory, Brain and Skill framework, and Lesson 10 added the BRAIN framework of Base, Rules, Actions, Intelligence and Navigation. They wrote their own quality-standards document as the Rules layer and built two custom Claude Skills to read and extract each newsletter. Every new source now moves through the same four stages, ending in a written log that carries the work across sessions.
“Real data needs a live quality-control pass — direct API writes beat CSV re-import for anything you'll maintain.”
In their own words
After
Their pipeline reads a newsletter in French, Dutch or English, strips it to its science and feeds a five-table Airtable database of letters, conditions, remedies, sources and prescriptions. During the submission the agent scaled the database live, added peer-reviewed papers, fixed three data-integrity issues through the API and wrote a full chapter on Bacopa monnieri and cognitive ageing. They also tested it on two other named health writers, and it held up.
What it lets people do
The pipeline lets a researcher turn a pile of informal health newsletters into a searchable, linked body of knowledge. It writes chapters that rank evidence, use full Latin names, give dosage ranges rather than prescriptions, and treat newsletter claims as anecdote only.
What they learned
They learned that real data needs a live quality check, and that writing directly through the API beats re-importing files. Above all, they found that refusing to pad the count with promotional or invented content made the work stronger. In their words, working from real sources only was a feature, not a limitation.
“Working from real sources only turned out to be a feature, not a limitation.”
In their own words
See the work
Wetenschappelijke e-mail-extractie: phytotherapy knowledge pipeline was built during AI Mastery.
Read the full submission
Written 8 August 2026 Submitted by the storyteller
Every story is written by a Mindvalley student or by someone our teachers have worked with. How we collect and check stories
More from AI Mastery
