I Built an AI Tool to Explain Confusing Medical Bills — What I Learned Launching It

I’ve been helping friends and family understand their medical bills for years.

EOBs, adjustments, denials, insurance codes… even people who work in tech struggle to make sense of them. And up to 60% of medical bills contain errors, which makes this even worse for patients.

So I built something I wish existed earlier:

👉 DecodeMyForm AI
An AI tool that reads a real medical bill/EOB and explains it in simple, human language — what happened, what insurance paid, and what you actually owe.

I launched it this week and wanted to share some lessons from the build + launch process, in case it helps other founders working with AI + messy real-world data.

💡 Key Challenges (That I Underestimated)

1. Healthcare documents have no consistent format

Every provider and insurer uses a different layout.
OCR breaks unless you normalize the structure.

2. GPT hallucination is real

Without schema alignment + extraction cleanup, the model fills gaps with “best guesses.”
Accuracy only happened after:

OCR tuning

Field grouping

Layout detection

Post-processing constraints

3. The real user pain is not “What does this bill say?”

It’s: “What should I do next?”
People want practical next steps, not just translation.

4. Privacy was the #1 blocker for early users

Before accuracy. Before speed.
If users see anything that feels like “upload your medical data to the cloud,” they bail instantly.

5. Upload flow UX is a conversion killer

Even a small friction point = instant drop-off.
I redesigned the flow 4 times before launch.

🛠 Tech Stack (for those curious)

React (Vite)

Node/Express backend

Google Gemini + OCR pipeline

Firebase Auth (magic link)

Stripe for lifetime/annual upgrades

No data stored — everything processed in-memory

📈 Why I Posted This on Indie Hackers

I know a lot of founders here deal with:

OCR + unstructured data

AI pipelines

Consumer trust challenges

Regulated-adjacent domains

Real-world messy inputs

I’d love feedback from anyone who has worked on:

AI document processing

Healthcare/insurance UX

High-friction user onboarding

Pricing strategies for B2C AI tools

Anything you learned the hard way?
I’m actively collecting insights.

🔗 If You Want to Try It (or see the PH launch)

👉 DecodeMyForm AI on Product Hunt
https://www.producthunt.com/products/decodemyform-ai

No signup required to test the sample EOB.

🙌 Would love your feedback

If you’ve built something similar — even in a totally different industry — I’d love to learn what worked (and what didn’t).

What’s the biggest challenge you faced when processing unstructured documents with AI?