Location: San Francisco Bay Area

Job type: Full-time

Application is linked at bottom of page.

About us

Curio is the high-stakes competitive game company. Duper is Curio’s flagship social war strategy game with onchain wager, a game where Poker meets Risk with a twist of Diplomacy. Our team comes from top games such as Genshin Impact, Halo, and Pacific Rim, and we have raised 8M+ in Seed from Bain Capital and others. Duper Season 1 had 10k+ players, many of whom are describing Duper as “the most fun game in web3”. Our vision is to build the ultimate fan-powered esports platform around Duper’s core loop and IP.

Links:

At Curio, we are building a stellar team to shape the future of competitive gaming. Join us.

About the role

RL-based agents in games like League of Legends and Honor of Kings are critical for playing onboarding because they allow controlled difficulty setting and removal of matchmaking wait time. This is the same for Duper: Duper is not an easy game to master, and a game agent can drastically improve the onboarding experience if done well. Papers including DeepMind’s AlphaStar and OpenAI’s FIVE are good examples of this.

However, Duper has important differences with MOBA games: Duper is a game heavy in not only strategic gameplay, but also negotiation, deception, and alliance-building through natural language interactions. While the former is much more limited and discrete compared to MOBA, the latter makes it a complex challenge for pure RL approaches. In this regard, Duper is similar to an old board game, Diplomacy. Meta AI’s CICERO is a good example using RL+NLP.

Your role is to work on training and maintaining the Duper agent. Your work will likely involve RL and infrastructure engineering, data science, and LLM engineering. Lots of guidance will be provided by our ML tech lead and adviser who has led agent building at Riot. If successful, this effort will evolve into Curio’s broader AI strategy.

Responsibilities