Research Scientist, Post-Training — Video Generation

Rozgoo; United States; Hybrid; USD 150,000 - 230,000; Full time

Benefits offered by this job

Full health benefits

401k matching

Equity

Hybrid work model

Job summary

Pika seeks a Multimodal LLM Researcher to push the boundaries of real-time, multimodal AI across text, image, video, and audio. You will design novel architectures, distillation methods, and world models, while prototyping agentic systems that power scalable, interactive multimedia experiences.

You will collaborate with engineering and product teams, publish in top venues, and contribute to production-ready technologies. This is a hybrid role with HQ in Palo Alto, CA.

Qualifications

• 5+ years of research in LLM/ multimodal models (LLM, VLM, Audio LM) in academia or industry.
• First-author publications in top conferences/journals (e.g., NeurIPS, CVPR, ICML, ICCV, SIGGRAPH, Interspeech).
• Deep expertise in LLM, VLM, or Audio LM.
• Experience with autoregressive and diffusion models for real-time deployment.
• Hands-on experience curating and building large multimodal datasets.
• Experience deploying real-time systems and agentic orchestration infra.

Responsibilities

• Lead research on real-time multimodal generation (text, image, video, audio) and orchestration of agentic platforms.
• Design and prototype novel algorithms for high-fidelity multimodal synthesis and interactive experiences.
• Focus on real-time model inference across modalities.
• Work on diffusion model distillation and diffusion-based world models for multimodal apps.
• Train and finetune models (LLM, VLM, Audio LM) for real-time performance.
• Curate datasets for video, audio, cross-modal data.
• Collaborate with engineering/product to productionize research advancements.
• Publish results in top venues and communicate internally/externally.

Skills

Strong programming & prototyping

Research in LLM/VLM/Audio LM

Real-time systems development

Excellent communication & teamwork

Dataset curation & multimodal data

Python programming

PyTorch proficiency

TensorFlow proficiency

Education

PhD or MS in a relevant field

Tools

Python

PyTorch

TensorFlow

Job description

Multimodal LLM Researcher (MLLM)

About the Role

At Pika, we are pioneering next-generation creative infrastructure built around real-time, multimodal generation and intelligent, agentic platforms. We are seeking accomplished Multimodal LLM Researchers (LLM, VLM, and Audio LM) to drive forward our mission to make agentic real-time generative technology accessible, dynamic, and transformative for millions of creators. As a core member of our research team, you will be integral to designing and building foundational technologies, developing novel approaches for large multimodal language models (LLMs/VLMs/Audio LMs), and orchestrating intelligent agentic systems that power scalable, interactive multimedia experiences. You will collaborate closely with engineering and product teams, shaping the future of real-time creative platforms.

What You’ll Do

• Lead and contribute to research efforts focused on real-time, multimodal generation—including text, image, video, and audio synthesis—as well as orchestration of agentic platform infrastructure
• Design and prototype novel algorithms and architectures for high-fidelity, real-time multimodal synthesis and interactive experiences
• Focus on real-time aspects of model inference and synthesis across modalities
• Work on diffusion model distillation and/or develop diffusion-based world models for multimodal applications
• Train and finetune autoregressive and diffusion models in LLM, VLM, or Audio LM contexts with a focus on real-time performance
• Curate specific datasets, especially for video, audio, cross-modal, and sensory-rich data
• Collaborate with cross-functional teams to bring research advancements into production-ready technologies
• Publish work in top-tier conferences and journals; communicate research results internally and externally
• Stay at the cutting edge of real-time multimodal generative AI and agentic orchestration

What We’re Looking For

• 5+ years of relevant experience, including research during graduate studies, in large language models, vision-language models, audio language models, deep learning, or related fields
• Demonstrated impact as first author on major publications in top conferences or journals (e.g., NeurIPS, CVPR, ICML, ICCV, SIGGRAPH, Interspeech, etc.)
• Deep expertise in at least one area: language modeling (LLM), vision-language modeling (VLM), or audio language modeling (Audio LM)
• Strong experience with generative models, including autoregressive and diffusion models, and their real-time deployment
• Hands‑on experience curating, constructing, or augmenting large, high-quality multimodal datasets
• Experience developing and deploying real-time systems and/or agentic orchestration infrastructure
• Strong programming and prototyping skills (Python, PyTorch, TensorFlow, etc.)
• Passion for building creative tools and platforms that empower users
• Excellent communication and collaboration skills

What We Offer

• Competitive salary and substantial equity in a high-growth startup
• Full health benefits + 401k matching and more
• Collaborative, mission-driven team environment with major growth opportunities
• Flexible on-site/remote hybrid (HQ in Palo Alto, CA)

About Pika

Pika empowers creators by building state-of-the-art agentic and multimedia platforms. Our vision is to break down technical barriers to creativity, making real-time generative and intelligent orchestration accessible to all. Join us and shape the next evolution of creative technology! If you are a leading researcher excited by real-time multimodal AI and agentic platforms, we want to hear from you.