Georg M. Goerg

Building decision guidance for claims examiners

Mon, 01 Jan 0001 00:00:00 +0000

EvolutionIQ’s AI guidance for frontline claims examiners.

Opening the workers' comp market with three new products

Mon, 01 Jan 0001 00:00:00 +0000

Three new workers’ compensation products, each with a full engineering team.

The LLM platform that cut building an extractor from a week to under an hour

Mon, 01 Jan 0001 00:00:00 +0000

An NLP/LLM platform for EvolutionIQ’s product teams.

Alerting and anomaly detection at Google

Mon, 01 Jan 0001 00:00:00 +0000

Statistical machine learning for alerting and anomaly detection at Google.

From a LinkedIn reply to a NeurIPS workshop paper

Fri, 30 Oct 2026 13:00:00 +0000

Igor Shuryak asked a question about radiotherapy dosing on LinkedIn. I replied, we ended up working on it together, and the paper is out at a NeurIPS 2026 workshop.

What all the pangrams say about how people play

Wed, 28 Oct 2026 12:00:00 +0000

I went through the public Pangramble leaderboard. Everyone reaches for QUICK, FOX and JAZZ, and the shortest sentences use made-up words.

Ranking the members of an LLM council

Mon, 01 Dec 2025 12:00:00 +0000

A fork of Karpathy’s llm-council that ranks the council members with a Plackett-Luce model and lets a human join.

How I built Pangramble

Sun, 20 Oct 2024 12:00:00 +0000

A daily pangram game, built in plain TypeScript to learn the language.

About Me

Mon, 01 Jan 0001 00:00:00 +0000

Research · Engineering · Leadership

Research

Mon, 01 Jan 0001 00:00:00 +0000

Featured

Setting a Dose versus Observing One: Continuous Dose-Response Estimation under Right-Censoring Georg M. Goerg and Igor Shuryak (Columbia University). Accepted at the NeurIPS 2026 Workshop on World Models for High-Stakes Health (WMHS).

PSPS-CS is a deep mixture-of-experts model that estimates whole dose-response curves for continuous treatments, such as a radiotherapy dose, when outcomes are right-censored survival times and confounding is strong. In simulations it was competitive with five other estimators. On a real head and neck cancer cohort the data support no dose contrast and no change in current clinical practice; the model is meant as a complementary tool for hypothesis generation.

Software

Mon, 01 Jan 0001 00:00:00 +0000

Python

Install these from GitHub; packages with similar names on PyPI belong to other people.

pyforeca: Forecastable Component Analysis.
pypsps: Predictive State Propensity Subclassification, causal deep learning in Keras.
torchlambertw: Lambert W x F distributions in PyTorch.
pylambertw: scikit-learn interface to Gaussianize heavy-tailed, skewed data.
pypress: Predictive State Smoothing (PRESS), non-parametric regression in Keras.

R

The R packages ForeCA and LambertW are on CRAN.