Judit Kisistók

About

My career has taken me from cancer research to fashion tech, and everywhere I've worked, I've seen the same patterns: teams either fire-fighting and shipping models without the explainability or observability work (making adoption and maintenance much harder), or endlessly perfecting models that never ship at all. That gap - between a model that works and one people actually trust enough to act on - is where I've spent most of my career.

When I'm not predicting the next pink camo jacket trend, you'll find me mentoring the next generation of devs at Hack Your Future Denmark or perfecting my knitting tension (both require surprising amounts of patience and problem solving).

Production ML Explainability Observability

Portfolio

Currently working on

Fashion Demand Forecasting

The NPI module turned out to be more popular than anyone expected - so now I'm making it even better (read: incredibly customizable).

D3 + React Data Visualization

A fun little project to experiment with interactive data visualization on the web.

LLMOps on Databricks

Building and deploying an agentic system that (hopefully) won't go off the rails. Focusing on evaluation, observability, and the unglamorous-but-essential side of LLMOps.

ML Engineering · Fashion Demand Forecasting

New Product Introduction Module

Turns out, predicting demand for products that don't exist yet is... complicated. I built the module that bridges the gap between what the model thinks will sell and what the buyer actually needs to make a decision.

Research · Nature

Tracking Early Lung Cancer Metastatic Dissemination in TRACERx

Can we use liquid biopsies to track early metastatic dissemination in lung cancer? This paper, which I co-authored with the TRACERx team, explores this question.

Read the paper →

Patents · EP3897605 & US20230039766

Combination Therapy for Cancer Treatment

This started out as a 2-person experimental project with an MD colleague. I built a proprietary ML model for identifying pharmaceutically suitable drug combinations, and we interpreted the results together as we went. This ended up in some patents, so I think we did well.

View the patent →

Open Source

ggAU

I got tired of customizing my R plots from scratch every single time, so I made a ggplot2 package with Aarhus University's color palettes and some sensible defaults. Surely gets you to publication-readiness faster.

View on GitHub →

Research · Academic

PhD Dissertation

Spent a few years investigating circulating tumor DNA - what it can tell us about cancer biology, and whether (and when) it's actually useful in the clinic. The answer seems to be: yes, but it's complicated.

Read my thesis →

Talks

Budapest Data + ML Forum

From Gut Feelings to Data-Driven Decisions

A talk about the gap between "our model is 94% accurate" and "our team actually uses this thing." Recognized as a Top 3 speaker for practical insight and technical depth, so I think it resonated.

View slides →

15th Current Topics in Bioinformatics · Berlin · May 2019

How Cell Simulation Can Help Precision Oncology

What happens when you use cell simulation to understand why some CLL patients respond to drugs and others don't? A case study covering the methodology, the surprises, and the results from an internal research project. (Bonus: at this point, I was the only non-Dr speaker on stage.)

Startup SAFARI 2019 · Budapest · Apr 2019

Cracking the Code of Cancer: Applying Machine Learning to Understand Patient Response

A behind-the-scenes look at a real research project, presented to a room full of founders and tech enthusiasts who'd never heard of chronic lymphocytic leukemia before.

Contact

I'm always excited to connect with fellow ML practitioners, anyone building something interesting / inspiring / delightfully weird, and people navigating squiggly career paths. So if that's you, holler 👋

Connect on LinkedIn → GitHub →