Verify the unverifiable

Soft skills, hard performance.

Madistii is an RL studio for domains you can't unit-test: style, voice, judgment, taste. We build verifiable, robust rewards for those skills, so an open-weight model can gain them without overfitting to a flat average, or losing existing capabilities on benchmarks and real-world tasks.

Flagship product

Misty Lossless Style Transfer for Open-Weight Models

Misty is Madistii's flagship product: give an open-weight model a specific voice, while keeping every bit of capability it already has.

Model Qwen/Qwen3-8B
Prompt

Explain why the sky is blue.

Style

Ernest HemingwayAuthor

Excerpt – A Moveable Feast

The tank wagons were painted brown and saffron color and in the moonlight when they worked the rue Cardinal Lemoine their wheeled, horse-drawn cylinders looked like like Braque paintings. No one emptied the Café des Amateurs though, and its yellowed poster stating the terms and penalties...

SFT Fluency ✓Accuracy ✗

The water is blue and the sky takes its color from below. On the grey days, the sea goes grey and the sky goes with it.

Naive RL Fluency ✗Accuracy ✓

Ah, the sun's light–white, and the air scatters its short blue wavelengths far more than the red, so the sky you see is blue.

Misty Fluency ✓Accuracy ✓

Sunlight carries every color at once. The air is full of small things far smaller than dust which catch the blue light and throw it across the whole sky, and all the other colors of the light much less so.

Portrait of Daehyun "David" Kim

Our Team

Founder-led research,First-principles thinking.

David Kim

Founding Engineer

David is a Founding Engineer at Madistii and a SWE at SAS. He studied at UNC–Chapel Hill, where he double-majored in Music and Data Science, and served as principal cellist of the UNC Symphony Orchestra.

At Madistii, David leads product direction and supports core research efforts in robust post-training; leveraging his extensive academic and professional background in deep reinforcement learning.