Founding AI Interpretability & Game Systems Lead / Technical Co-Founder — Neuro-Game

Founding AI Interpretability & Game Systems Lead / Technical Co-Founder — Neuro-Game™

San Francisco Bay Area · Hybrid / Remote · Equity (pre-seed)

What we're building

Neuro-Game decodes the patterns inside AI models and retrains the ones that shouldn't be there.

Our bet: games are a uniquely revealing way to probe a model. The human brain opens up when it plays, and we believe models do too. Under structured play, they show behavior and internal representations that standard evals miss. We see this as one layer in a stack of dedicated systems that keep powerful models in check.

The vision and framework are in place. Now we need the person who turns them into working science.

What you'll own

• Designing and building the games models play: rules, mechanics, scoring and environments that reveal how a model reasons, not just what it outputs

• Interpretability tooling: activation analysis, probing, sparse autoencoders, circuit-level investigation on open-weight models

• Linking game behavior to what's happening inside the model, then to targeted interventions (steering, fine-tuning, unlearning)

• Rigorous evaluation that shows whether our methods work, honestly and reproducibly

• Research direction, first hires and the technical roadmap

You might be a fit if

• You have hands-on mech interp or alignment experience (MATS, papers, open-source tools, or serious independent work)

• You've built games, simulations or RL/eval environments, and you think in mechanics and incentives

• You're fluent in PyTorch and comfortable working with transformer internals

• You'd rather publish a negative result than oversell a positive one

• The idea of reading a model's mind through play excites you

The deal

This is a founding role. Compensation is equity only until our first raise, then a market salary plus equity.

About me

I'm Fran Opris, MSc, founder and CEO of Neuro-Game. I come from neuroscience, where decoding the hidden patterns that drive behavior is the core problem. I bring the framework, the product vision and the go-to-market. I'm looking for the technical co-founder who can prove it on real models.