My First Attempt at a Medical AI with ELI5

Prologue — When Writing Became Experiment

During the Flame Glyph experiments, I pushed the boundaries further than I expected.

Not just symbols, not just encoded meaning —

but entire writing itself.

Curious what Flame Glyph really is?

👉 Here’s the story.

At one point, I even embedded a manuscript called The Covenant of Life into its structure.

It was never meant to be published. It was an experiment:

Could writing itself become part of a living system of symbols?

Quietly, another thought followed.

Maybe one day, this wouldn’t remain metaphor or art.

Maybe it could become something practical —

a medical SaaS, a tool that might help patients.

It felt naive. Unrealistic. Almost laughable.

But the idea wouldn’t leave me.

And then, as the saying goes:

“When your eyes are ready, you begin to see.”

That was the moment the impossible started to feel real.

1. When I Began to See

One day, while idly scrolling through the news, I stumbled on a press release from the Ministry of Science and ICT.

The headline read:

“Boost up AI 2025 — National AI Competition on Drug Discovery.”

I clicked.

What I saw was unlike anything I had imagined:

a real dataset from the Korea Chemical Bank —

1,681 measured interactions between compounds and the enzyme CYP3A4.

It was the very enzyme I had just learned metabolizes more than half of all known drugs.

The challenge was simple on paper but enormous in practice:

Build an AI model that could predict inhibition rates for 100 new compounds

with accuracy judged by RMSE and Pearson correlation.

That was the day my quiet thought stopped being hypothetical —

and began turning into work.

2. Facing the Unknown

I knew nothing about medicine.

Nothing about CYP3A4, the enzyme that metabolizes more than half of all known drugs.

I had never touched big data modeling.

Never trained a machine learning model.

Never fine-tuned anything beyond words on a page.

So I turned to AI.

I asked:

What should I learn — not only about CYP3A4, but also about the technical stack behind a real SaaS?

The answers came back like a syllabus:

Python for algorithm prototyping

APIs and backend integration

Basics of pharmacology and drug metabolism

It was overwhelming.

But strangely —

I wasn’t afraid.

3. The Magic Spell: ELI5

In those early days, every search felt like hitting a wall.

Medical jargon. Machine learning papers. Endless acronyms.

Then I stumbled on a magic word: ELI5.

“ELI5” stands for Explain Like I’m 5.

Type it before any question, and suddenly AI would rewrite complexity into simple truths.

Through ELI5, I learned what CYP3A4 does inside the liver.

Through ELI5, I finally understood what an API actually is, and how different pieces of a SaaS stack fit together.

It wasn’t just a hack.

It was a spell.

👉 A spell that turned fear into curiosity —

and curiosity into progress.

4. Two Weeks of Pure Learning

I began where anyone could: GitHub.

I opened repositories on CYP3A4 and pharmacology,

asking AI what each piece of code meant — and why it mattered.

Repo after repo, the patterns slowly appeared.

I added arXiv papers to the mix.

For two weeks, I did little else but study.

Line by line, concept by concept, things became clearer.

It wasn’t easy.

I learned. I unlearned. And I learned again.

I began with nothing, building layer by layer, trying to take in as much as I could.

5. Week Three — Building ARR-MEDIC

By the third week, I felt ready to try.

With AI’s help, I coded a simple algorithm.

I ran mock simulations.

I combined everything I had learned — until a prototype came to life:

👉 ARR-MEDIC, an AI model to predict CYP3A4 inhibition.

It wasn’t perfect.

So I asked again and again: What’s missing? How can this be better?

AI answered. I refined.

Step by step, ARR-MEDIC began to improve.

6. The Submission That Never Happened

At last, the system was ready.

But there was one problem: I wasn’t in Korea.

I was living in Thailand with my Thai wife, our son and daughter — holding together our small family.

Still, I wanted to try.

So I wrote to the organizers:

“Dear team,
I independently developed a medical AI platform, ARR-MEDIC.
It predicts CYP3A4 inhibition, calculates clinical risks, embeds an ethics protocol, and connects with DrugBank.
This is not just a contest entry — it is my attempt to show what’s possible in medical AI.”

The reply was swift, and final:

“This is not a submission-based competition. Please refer to the official page.”

But the page was blocked from abroad.

And just like that, my entry was over before it even began.

7. What I Gained Anyway

It hurt.

I had spent a month — sleeping three or four hours a night — pouring everything into ARR-MEDIC.

And yet, I couldn’t even submit.

But I didn’t walk away empty-handed.

I learned what CYP3A4 is, and why it matters in drug development.

I built my first medical AI prototype from nothing.

I discovered how GitHub, arXiv, and AI itself could become my teachers.

Most of all, I proved something to myself:

that even without credentials, I could enter a field with nothing but curiosity — and one spell: ELI5.

8. What I Built

After the failed submission, I kept building anyway.

That work became ARR-MEDIC — a open-source on GitHub project for predicting CYP3A4 inhibition.

Its strengths are clear:

Focused: it concentrates only on CYP3A4, making the purpose easy to understand.

Accessible: available with a Hugging Face demo, so anyone can try it without heavy setup.

Lightweight: runs quickly, doesn’t require expensive resources, and is beginner-friendly.

Educational: designed as a tool to learn from, with clear documentation and examples.

It may not be the most powerful model, but it lowers the barrier.

It offers an entry point for students, researchers, and curious outsiders —

the same way I began.

👉 ARR-MEDIC on GitHub: please check the first comment.

9. What I Learned

In one month, I realized something simple but profound.

We live in a time when, if you truly want it, you can learn, you can build, and you can even create your own opportunities.

For me, the key was one word: ELI5.

It was the spell that broke down walls, turned confusion into clarity, and made the impossible approachable.

And I saw this truth clearly:

you don’t need to graduate from medical school,

you don’t need to be a senior developer with years of experience.

If you have urgency, if you have the will to learn,

then this is the era where you can make something real.

That is what ARR-MEDIC ultimately gave me —

not just a prototype, but the conviction that building is possible.

Epilogue — From Failure to Vision

So what am I doing now?

I’m refining ARR-MEDIC into a working SaaS prototype —

strengthening its reasoning, adding safeguards, and testing it against real-world benchmarks. Ahead still lies full EMR/FHIR integration.

That is where the path now leads

Every day, I still whisper the same magic spell: “ELI5.”

And with it, I keep dreaming —

👉 that one day, the medical AI SaaS I built might help save lives.

Maybe that is the real promise of ELI5 —

not just clarity, but possibility.