Our Company

We’re building the platform for worldbuilders, storytellers, and game devs to create and publish any ‘what if’ they can think of. So far, most of the 6k+ user published games on our site are historical– what if Rome never fell? what if you could go back to 2016 and start over?– but we’re growing into more genres by the day.

Our 50k+ daily active users play different games by writing free text actions and chats with other countries. AI then powers the worlds’ response (make an account to try it yourself, or demo here). We’ve seen nearly 43 million rounds played and are growing quickly.

We’re backed by Y Combinator (W26), Bessemer, Pace Capital, Z Fellows, and more. Join us to help build the future of interactive entertainment.

The Role

We need a founding level AI/ML developer to work in-person full-time in San Francisco. Your north star metric will be to improve user-preference win-rates over off-the-shelf options with the same inference budget.

Some fun challenges we’re thinking about, and you’ll be working on:

We collect and publish user preference data on which models give the best responses, and have a working internal eval system, but there’s a lot more to be done here.
We need to build and run the infrastructure needed to rigorously tailor harnesses and prompts to individual AI models.
Training domain-specific models to close or even eliminate the gap between open and closed models in their weight class at playing Pax.
Using optimal caching strategies to reduce costs associated with closed source models.
Improving closed source models’ performance by training tuned endpoints.
Evaluating and improving embedding and reranker performance.
Working with teams at Chutes, OpenRouter, CanopyWave, and more.
Enabling entirely new user experiences based on upcoming world models (3D!).

We’re moving really fast, and want someone on our team who’s prepared for that. To set some expectations up front, you should be prepared for:

Longer hours (50-60/week) in person
Fast pace and strong ability to quickly switch tasks/priorities
Frequently learning new things (must be comfortable tackling truly unknown problems from scratch)
High degree of ownership expected– in return, high degree of trust will be given
Working directly with cofounders
Using trillions of tokens of prompt and response logs from millions of gameplay trajectories to create an eval & training system which produces compounding improvements.

Your Qualifications

What we care most about:

Extremely high slope. Great learners are way more important to us than people who have any particular experience.
Personal drive. We love demos and side projects– please submit any personal projects/demos you may have. Past startup experience is also great.
Passion for product. Most of our players have discovered Pax Historia organically (friends, youtube reviews) and have stuck around because they truly care about the platform we’re building. That’s why we want every one of our employees to care deeply about our product too.

Experience we expect:

You have shipped ML systems to real users and operated them in production.
You have made explicit cost/quality tradeoffs in deployed systems.
You have designed, critiqued, or iterated on evaluation frameworks and understand their failure modes.

Experience we’d get extra excited about:

Experience with preference modeling, pairwise ranking, or human-in-the-loop evaluation systems.
Background in games, simulations, storytelling systems, or other domains where qualitative judgment matters.
Experience operating systems at high request volume.
Prior work at an early-stage startup or as a founding engineer.

---

This job listing is for a W-2 employee opportunity. We are unfortunately unable to sponsor visas (other than O1) at this time. Pax Historia is an equal opportunity employer and does not discriminate on the basis of race, color, religion, sex, gender identity or expression, sexual orientation, national origin, age, disability, or veteran status.

Pax Historia

Open roles(3)