About this role
Summary: Design and build next-generation agentic systems leveraging visual reasoning and VLMs to create personalized user experiences. The role involves prototyping agentic architectures, evaluating research, and collaborating with ML researchers and engineers to deploy robust production-grade systems.
We're looking for an Agentic AI Researcher to join our team’s venture - Designing and building the next generation agentic systems enabling personalized experiences for our users. You'll build agentic systems that unlock the full potential of VLMs, turning strong visual reasoning into experiences tailored to each domain, deploying latest research into high-quality, robust production-grade systems.
Description
- Design, prototype, and ship agentic architectures that operate over user's visual data.
Track, evaluate and adapt the latest research and frameworks, turning SOTA designs into production-grade systems.
- Explore new approaches to visual perception within agentic loops, for tasks of video understanding, spatio-temporal grounding, efficient visual tokenization, and more.
- Investigate the tradeoffs between compact on-device models and frontier cloud VLMs to design a hybrid routing system for latency, privacy, and quality.
- Collaborate with ML researchers and engineers on the underlying generative models that agent tools depend on.
- Partner with product and design to translate open-ended creative requests into reliable, testable agent behaviors.
Minimum qualifications
MSc in Computer Science or a related field
Proficiency in Python, PyTorch, and other modern ML frameworks
Prior research or work experience with agentic multimodal LLM systems - self-evolving agentic system, multi-agent system orchestration design, RL for agents, reflection-action agentic loops, agentic alignment and safety.
Motivation to work and collaborate within a friendly, innovative and cooperative team
Preferred qualifications
Strong background in machine learning research or applied research, with hands-on experience building systems powered by Vision, LLMs or multimodal models.
Proven experience in product-focused applicative research in the industry
Research track record (publications in NeurIPS, CVPR, ICCV, ECCV, etc.)