How I’m Using AI to Build a Complete St. Alphonsus Liguori Publishing Program

Hi everyone,

Pardon me, but I am going to allow AI to explain this so its more understandable for you.

For years, I have published public-domain books through Amazon and other publishing platforms.

Today, I have more than 1,700 titles in my publishing catalog.

But lately, the way I approach publishing has begun to change dramatically.

Instead of simply finding an old book, preparing it, and publishing it, I am experimenting with something much larger:

Can AI help me systematically map the works of an important historical author, locate the best source editions, identify what I have already published, determine what is missing, produce those missing books, and eventually create one of the most complete collections of that author's works available?

That is exactly what my AI team and I are now attempting with St. Alphonsus Liguori.

And if the system works, Liguori is only the beginning.

Why St. Alphonsus Liguori?

St. Alphonsus Liguori was an extraordinarily prolific Catholic author.

Over the years, I have already published a number of his works. But until recently, I did not have a reliable answer to a surprisingly basic question:

Exactly how much of his work do I already have—and how much am I missing?

That question led us to a much larger project.

Instead of looking for one more Liguori book to publish, we decided to try to identify his entire English-language publishing corpus.

The long-term goal is ambitious.

I want to determine whether we can eventually offer something approaching 99% of the viable English works of St. Alphonsus Liguori in a professionally produced collection.

That changes the business model considerably.

Instead of competing one book at a time, I could eventually offer readers something much more valuable:

A definitive St. Alphonsus Liguori collection.

Someone deeply interested in Liguori does not have to search dozens of websites, publishers, archives, and used-book listings.

We do the work of finding, preparing, organizing, and publishing the collection.

But getting there requires much more than asking AI:

“Find all the books written by St. Alphonsus Liguori.”

That would be far too unreliable.

We are building the project in stages.

Step 1: Build the Complete Bibliography

The first task is discovery.

We use AI-assisted research to search bibliographies, library catalogs, digital archives, historical publishing records, and other sources for works attributed to Liguori.

At this point we are deliberately trying to cast a wide net.

We want titles.

Alternate titles.

Translations.

Volumes.

Abbreviated editions.

Collections.

Individual works that may later have been published inside larger collections.

The objective is not yet to decide what should be published.

The objective is:

Find the universe of possibilities first.

That distinction is important.

AI research can generate candidates very quickly, but a candidate is not automatically a legitimate new book.

Step 2: Determine Whether Two Titles Are Actually the Same Work

This is one of the biggest problems when working with historical books.

A single work may have been published under several English titles.

One translator may use one title.

Another publisher may use another.

An abbreviated edition may use a third.

Without careful research, it would be very easy to look at three titles and conclude that they represent three different books.

They may actually be the same work.

So our next step is identity normalization.

We compare titles, authorship, contents, publication information, title pages, translators, tables of contents, and other evidence.

The AI is instructed not to assume that a slightly different title represents another work.

When the evidence is insufficient, the book remains unresolved.

That is an important principle in the system:

Uncertainty is allowed. Guessing is not.

Step 3: Identify the English Editions

Once we know that a work exists, another question appears:

Was it translated into English?

If so:

Which translation?

When was it published?

Who translated it?

Was it complete?

Was it an abridgment?

Was it part of another volume?

Different English editions can be substantially different.

For that reason, simply finding the name of a book in a library catalog is not enough.

Where possible, we want evidence from the actual printed book—especially title pages, publication pages, prefaces, tables of contents, and the text itself.

This begins to turn a simple book list into something much more valuable:

A publishing intelligence database.

Step 4: Compare the Corpus Against My Existing Publishing Catalog

Now we ask:

Which of these books have I already published?

Remember, I already have more than 1,700 titles.

Some books may already exist under slightly different titles.

Some may have been published years ago.

Some may exist as a Kindle edition but not a paperback.

Others might have a Regular Edition but no Large Print edition.

Therefore, the AI system has to reconcile the Liguori research against my existing catalog rather than assuming every discovered book represents a new publishing opportunity.

Ultimately, every Liguori work should fall into categories such as:

• already published;
• partially published;
• missing;
• requires additional research;
• source unavailable or unsuitable;
• possible new format opportunity.

This is where the project starts becoming commercially interesting.

We are not simply discovering books.

We are identifying gaps.

Step 5: Locate the Best Source Copy

Once we identify a missing work, we still do not immediately publish it.

First we need a source.

There may be copies available from places such as Internet Archive, Google Books, university libraries, or other historical archives.

But not every scan is usable.

One scan might contain missing pages.

Another may have poor photographs.

Another might have terrible OCR.

Another may be an abridged edition when we are looking for the complete work.

So the AI searches for candidate source editions.

Then we qualify them.

Step 6: Inspect the Source Before Production

This is one area where I believe our process has become much more sophisticated.

We do not want an AI system blindly processing a 400-page PDF just because the filename looks correct.

The source has to pass inspection.

We may check:

• the printed title page;
• author attribution;
• publication date;
• edition information;
• completeness;
• beginning and ending pages;
• missing or damaged pages;
• readability;
• OCR quality;
• images or illustrations;
• footnotes;
• page count;
• production complexity.

Claude/Cowork and other AI tools can assist with this type of qualification and visual inspection.

If the evidence does not establish what we need to know, the title remains on hold.

Again:

No guessing simply to keep the production line moving.

Step 7: Decide Whether the Book Is Worth Producing

Just because we can publish a book does not automatically mean we should.

Before spending time producing it, we can investigate the current market.

Is the title already readily available?

How many competing editions exist?

Are those editions professionally produced?

Is there a Kindle edition?

Paperback?

Large Print?

Could our version provide something useful that the market currently lacks?

AI can help perform this research far faster than I could manually inspect hundreds of titles myself.

But market research still does not make the final decision.

I do.

That remains one of the fundamental rules of the system.

AI researches.

AI organizes.

AI identifies possibilities.

AI prepares evidence.

The CEO makes the publishing decision.

And in my little publishing empire, that CEO is still me. :-)

Step 8: Design the Product

A selected title can potentially become more than one product.

Depending on the book, we may produce:

Regular Edition Paperback

Large Print Paperback

Kindle Edition

Hardcover

and eventually make appropriate digital editions available through one of my collections.

Large Print has become an especially important part of my publishing strategy when the book is suitable for it.

The decision is made title by title rather than automatically producing every possible format.

Step 9: AI-Assisted Transcription and Verification

Now the actual book-production work begins.

This is another place where AI has dramatically changed what is possible.

We have already tested this with Liguori's The Holy Mass.

The source was a nineteenth-century scan of more than 470 pages.

The workflow involved AI-assisted transcription of the book, followed by page-by-page checking against the original images.

Hundreds of notes also had to be handled.

Errors and uncertain readings were logged instead of silently invented.

The goal is not merely:

Create text that looks reasonable.

The goal is:

Create text that can be traced back to the source.

That distinction is enormous.

We are gradually moving toward a production line in which one AI can perform transcription, another can perform verification or anomaly detection, and questionable passages can be escalated for human review.

Step 10: Turn the Verified Text Into a Professional Book

Once the text has passed the transcription and QA stages, it still has to become an actual book.

That means typography.

Page size.

Margins.

Styles.

Headers.

Front matter.

Chapter formatting.

Large Print formatting when appropriate.

PDF export.

Final quality control.

We are now experimenting with ways for AI to assist with the Adobe/InDesign side of this workflow as well.

The dream is not to eliminate human oversight.

It is to eliminate repetitive human labor.

I should not have to manually perform hundreds of identical formatting operations if software can reliably perform them from rules I have already approved.

Step 11: Publish the Missing Titles

After final review, approved books can enter my normal publishing process.

That currently centers heavily on Amazon KDP, with additional opportunities through IngramSpark and my own publishing properties.

But something important has changed.

We are no longer randomly adding another book to the catalog.

Every new title moves us closer to completing the Liguori corpus.

There is now a destination.

Step 12: Build the Complete St. Alphonsus Liguori Collection

This is the part I find especially exciting.

Suppose our research eventually identifies the viable English Liguori corpus.

Suppose we already own 60%, 70%, or 80% of it.

We can systematically produce the missing books.

Eventually we may reach 90%.

Then 95%.

And perhaps something approaching 99%.

At that point I am no longer simply selling individual public-domain books.

I may be able to offer:

The St. Alphonsus Liguori Collection

A reader interested specifically in Liguori could obtain almost everything in one organized collection rather than purchasing dozens of books separately.

That collection could justify a premium price while still providing tremendous savings compared with buying every title individually.

This is one reason I think author-specific collections may ultimately become an important part of my publishing business.

A broad library appeals to a broad audience.

But someone intensely interested in one particular author may place considerably more value on a nearly complete collection of that author's writings.

Then We Do It Again

This is where AI changes the economics of the entire idea.

Everything we are learning with Liguori can become a repeatable system.

The next author does not require inventing the process again.

We already have the structure:

Discover → Identify → Verify → Reconcile → Source → Qualify → Research → Produce → Publish → Package

After St. Alphonsus Liguori, the same model can be applied to another major author.

One candidate already on my roadmap is St. Thérèse of Lisieux.

Then St. Robert Bellarmine.

And after that?

There are centuries of important public-domain authors whose works have never been systematically organized for today's readers.

Instead of thinking title by title, AI allows me to begin thinking author by author.

What AI Is Really Giving Me

The biggest benefit is not that AI can write faster than I can.

That is actually one of the least interesting parts of this experiment.

AI gives me something much more valuable:

Scale.

A human publisher could spend weeks researching an author's bibliography.

Then more weeks locating editions.

Then weeks comparing title variations.

Then months transcribing and checking books.

Then still more time performing market research and preparing production files.

With multiple AI systems working specialized jobs, much of that work can happen dramatically faster.

My role begins moving upward.

Instead of personally doing every task, I define the rules, review exceptions, make decisions, approve releases, and determine the strategy.

In other words, I am beginning to operate less like a person publishing individual books...

and more like the CEO of a small AI-assisted publishing company.

But AI Does Not Get the Final Vote

There is another lesson I have learned repeatedly.

AI can be incredibly productive.

It can also be confidently wrong.

That is why our system contains checkpoints.

Evidence matters.

Source pages matter.

Printed title pages matter.

Unresolved questions remain unresolved.

An AI finding something on the Internet does not automatically make it true.

And an AI recommendation does not automatically become a publishing decision.

I want the machines doing the work they are good at while preserving human judgment where it matters.

The Bigger Experiment

I do not yet know exactly where this will lead.

That is part of what makes it interesting.

We are building the Liguori system while using it.

Every difficult book teaches us something.

Every transcription problem improves the workflow.

Every failed source teaches us another qualification rule.

Every repetitive task becomes a candidate for automation.

Eventually the real intellectual property may not simply be the books.

It may be the system that repeatedly turns enormous historical author archives into organized, verified, professionally published collections.

That is a very different publishing business from the one I started with.

Years ago, publishing one book seemed like an accomplishment.

Today I have more than 1,700 titles.

Now AI has me asking a completely different question:

Instead of publishing the next book, could we systematically rebuild an author's entire body of work?

St. Alphonsus Liguori is where we are finding out.

And if we succeed, we already know what comes next.

We do it again.

To Our Success!

Mel Waller