Our Innovation team builds adversarial RL environments that train the worlds most advanced AI models to be safer. Our customers are the leading frontier AI labs, who use these environments for post-training reinforcement learning and safety evaluation. This is the bleeding edge of AI safety technology: the environments you build will directly shape how next-generation models learn to resist adversarial attacks.
Were looking for a Principal software engineer to own the RL Gym platform end-to-end: from architecting multi-site web environments that simulate real-world attack surfaces, to optimizing our in-house orchestration harness (AgenticVerse) for high-performance delivery into customer training pipelines. This is a builder role. Youll lead a small team (including a dedicated web environments engineer), operating with high autonomy, moving fast from concept to working prototype to production system. Youll interact directly with customer engineering teams to understand their infrastructure constraints and deliver environments that meet their scale and reliability requirements. Why this role
This is one of the few roles in the industry where your code directly influences how the next generation of AI models are trained. Youll be at the center of advancing AI safety, building systems that the worlds top labs depend on to make their models more robust. The work is technically deep, the problem space is genuinely novel, and the field is moving faster than any team can keep up with alone. Theres no playbook. Youll write it.
What youll do:
Platform & performance
* Own and evolve AgenticVerse, our in-house orchestration harness that provisions and manages RL environments at scale. Focus on performance: low-latency provisioning, high concurrency, minimal overhead per environment instance
* Design and build isolated, reproducible web environments using Firecracker microVMs or Docker containers
* Architect multi-site scenarios (3-4 interconnected web applications per task) with rich interactions: drag-and-drop, file uploads, authentication flows, LLM-in-the-loop components
* Implement deterministic verifiers that evaluate agent behavior with zero ambiguity Customer delivery
* Work directly with engineering teams at leading AI labs to integrate RL Gym environments into their training and evaluation pipelines
* Translate customer specs into working environments, iterating rapidly on feedback
* Own the technical relationship: SLAs, API contracts, integration architecture
* Adapt environment delivery formats to cus tomer infrastructure ( Real-Time API calls vs. offline batch, managed vs. raw artifacts)
* Build customer-facing UIs when needed (dashboards, environment configuration portals, monitoring interfaces) Rapid prototyping
* Take ambiguous problem descriptions and produce working prototypes within days, not weeks
* Validate new environment types, interaction patterns, and verifier approaches quickly
* Build internal tooling that accelerates scenario authoring and testing
About us:
A trust, safety, and security company built for the AI era. We safeguard the communicative technologies people use to create, collaborate, and interact-whether with each other or with machines. In a world where AI has fundamentally changed the nature of risk, we provide end-to-end coverage across the entire AI lifecycle. We support frontier model labs, enterprises, and UGC platforms with a comprehensive suite of solutions: from model hardening evaluations and pre-deployment red-teaming to runtime guardrails and ongoing drift detection.
Requirements: Must have
* 8+ years of software engineering experience, with a track record of building production systems from zero
* Deep expertise in infrastructure: Linux, containers (Docker), VMs (Firecracker or similar), networking, cloud platforms (AWS strongly preferred)
* Strong Python skills and comfort with async/co
This position is open to all candidates.