← Back to search

Nitrode

Data for reasoning agents in dynamic environments

Y CombinatorWinter 2025ActiveB2BArtificial IntelligenceMachine LearningB2BData Engineering

What they do

Problem

LLMs and AI agents trained on static data struggle with spatial-temporal understanding, memory, causality, and multi-step reasoning in dynamic environments.

Solution

Nitrode builds fully specified game environments that generate ground-truth state-transition data to train and evaluate reasoning agents on spatial and temporal dynamics.

Who it's for

AI research teams and developers building or evaluating LLMs and autonomous agents.

Founders

Ben Kim

Co-Founder

Co-founder at Nitrode. Infosci @ Cornell. Previously Data and VC/Corporate @ Wilson Sonsini.

Sejoon Chang

Co-Founder

Co-founder at Nitrode. CS & Product Design @ Stanford. Top 500 GeoGuessr globally in moving and builder of unnecessarily ambitious Minecraft cities.

Brian La

Co-Founder

Co-founder at Nitrode. CS @ Cornell. Previously Data Scientist @ Nexon

Open roles 3

Member of Technical Staff (Post-Training)

San Francisco, CA, US · Full-time · $150K - $350K · 1.00% - 3.00%

Member of Technical Staff (Evals)

San Francisco, CA, US · Full-time · $120K - $250K · 1.00% - 3.00%

Full-Stack Engineer

San Francisco, CA, US · Full-time · $110K - $160K · 0.50% - 2.00%

Company timeline

Founded2024
YC batchWinter 2025
StatusActive
Team size4 people
LocationSan Francisco, CA, USA

Similar startups

Post-training data to teach models document work

Training data for frontier AI labs

Expert STEM reasoning data for frontier AI labs.

Audio and Enterprise Human Data for AI

Training Data for Recursive Self-Improvement

Audio & Video Data