Vlad Miftodi

Co-owner/CTO @ RevUp Dental. Building production LLM data pipelines on healthcare calls. The model is the easy part; the data is the job.

[Vlad Miftodi @vladmif]

AI will write the code the second you ask. It can't tell you who it's for, what it has to do, or how you'll know it worked. That part is yours.

So before you build, answer the questions you keep skipping. AI didn't make them optional. It just took away your excuse.

[Vlad Miftodi @vladmif]

The most important work I ever did on our product happened at my parents' house, over Christmas, with a table of post-its and a lot of Kinder.

Not a single line of code.

A thread on what building an MVP is actually about:

[Vlad Miftodi @vladmif]

I think about this constantly now, because building got easy. People buy a subscription, open Claude, and have an MVP by the weekend.

A lot of them are building the wrong thing, beautifully, at record speed.

The tool got faster. The thinking didn't.

[Vlad Miftodi @vladmif]

The whole arc: prove it by hand, then let the machine take over. But only because you finally know what right looks like.

[Vlad Miftodi @vladmif]

How to reach product-market fit, by someone who took the slow road.

Our product started as an Excel spreadsheet. Here's what the long way taught me.

[Vlad Miftodi @vladmif]

Lesson 3: The manual work was never wasted.

Manual work costs you operations. AI work costs you validation. And you can't validate against nothing.

When we validated our AI, we had 500,000+ cleaned, human-reviewed records behind us. No answer key, no AI you can trust.

[Vlad Miftodi @vladmif]

@DanielSmidstrup Dentists pay to make the phone ring, then lose patients at the front desk and never know it.
Missed calls. Answered calls that never book. No follow-up.
We score every call, show what was lost and why, drive the follow-up to win it back, and prove the revenue recovered.

[Vlad Miftodi @vladmif]

@clarefinds I've been building for quite some time https://t.co/1AabGHiczK which helps dentists with end-to-end marketing

[Vlad Miftodi @vladmif]

A take from someone shipping a real LLM product: the model is the easy part.

The job is turning messy, real-world data into something a model can classify reliably, at volume, under HIPAA.

Going to document the unglamorous version, with numbers.

[Vlad Miftodi @vladmif]

@Stramosinft @UntoldFestival @missRuxandraD mergem la untold?

[Vlad Miftodi @vladmif]

@vladUngu @UoYOpenRes @asmasonomics Congrats. Know you’ve been hard at work on this one

[Vlad Ungureanu @vladUngu]

I am very excited about this tool which aims to make it easier to visualize gene expression data across datasets, Differential Expression Analysis and UMAP. Thanks @UoYOpenRes for the award and @asmasonomics for all the support