Multimodal ML Engineer

Responsibilities: Extend Runsphere’s runtime security coverage beyond text to the multimodal inputs and outputs used by modern AI agents.

• Research and build evaluations for image, audio, video, and mixed-modality model behavior.
• Create datasets and test cases for unsafe content, instruction conflicts, leakage, and policy bypass across modalities.
• Develop model pipelines and classifiers that can run reliably in customer workflows.
• Measure how multimodal models interpret context, instructions, and tool requests.
• Turn failure modes into regression tests, policy signals, and product requirements.
• Work closely with research, infrastructure, and product teams to ship practical controls.

Ideal candidate: You have experience with multimodal ML, computer vision, audio, video, or multimodal LLMs, plus strong Python and ML engineering skills. You can design experiments, build data pipelines, and evaluate models under adversarial or ambiguous inputs. You care about real-world safety and can explain results to engineers and product teams. Experience with agents, model serving, or AI security is a plus.