We are seeking a highly motivated Machine Learning Engineer to join our core research and development team, focused on video understanding and segmentation. In this role, you will build the systems that let us search, decompose, and describe massive volumes of egocentric and human-robot video at scale — turning raw, unstructured footage into structured, searchable, and richly annotated training data. You will work across video/image embedding models, LLM-based video understanding, and agentic pipelines that orchestrate multiple models into end-to-end workflows. This is a foundational role that directly shapes the data quality and scalability of our entire training data platform.
Responsibilities
Build and optimize video/image embedding pipelines using CLIP-style and other vision-language embedding models to power large-scale, multi-modal video search and retrieval.
Develop LLM-based video understanding systems for semantic indexing, summarization, and question-answering over long-form egocentric and third-person video.
Design and implement instruction-level and action-level video chunking/segmentation algorithms that decompose long videos into structured, temporally-aligned clips.
Build automated video captioning systems that combine vision-language models and LLMs to produce fine-grained, temporally-grounded descriptions of actions and scenes.
Architect agentic systems and orchestration pipelines that chain embedding, captioning, retrieval, and LLM reasoning steps into reliable, end-to-end video understanding workflows.
Develop and scale video search infrastructure (vector indexing, retrieval, ranking) to support semantic and multi-modal queries over millions of video clips.
Collaborate with annotation, data engineering, and robotics teams to integrate video understanding outputs into downstream training pipelines for embodied AI and robot learning.
Evaluate and benchmark embedding models, LLMs, and agentic frameworks against production needs; track frontier research and bring relevant techniques into the platform.
Contribute to internal tooling, documentation, patents, and open-source initiatives where applicable.
Mentor junior engineers and interns, and help shape the long-term technical roadmap for video understanding.
Minimum Qualifications
MS or PhD in Computer Science, Electrical Engineering, or a related technical field, or equivalent practical experience.
3+ years of hands-on experience in computer vision or multi-modal machine learning, with direct experience in video understanding tasks.
Strong proficiency in Python and PyTorch, with solid software engineering fundamentals.
Hands-on experience with CLIP or similar vision-language/video embedding models for retrieval or representation learning.
Experience building or fine-tuning LLM-based systems for video/image understanding (e.g., captioning, video QA, summarization).
Familiarity with agentic system design — tool use, multi-step reasoning, and orchestration frameworks (e.g., LangChain, LlamaIndex, or custom agent loops).
Experience working with large-scale video data pipelines and vector search/retrieval infrastructure (e.g., FAISS, Milvus, or equivalent).
Preferred Qualifications
PhD with a research focus in video understanding, multi-modal learning, or vision-language models.
Experience with temporal action segmentation, action localization, or instruction-level video chunking algorithms.
Experience working with egocentric video datasets or head-mounted-device (HMD) captured data.
Track record of deploying production-scale video search or retrieval systems.
Experience integrating foundation or vision-language models (e.g., CLIP, VideoCLIP, RT-1/VLA variants) into perception or decision-making pipelines.
Publications in top-tier computer vision or ML venues (e.g., CVPR, ICCV, ECCV, NeurIPS, ICLR, etc).
Experience with humanoid robotics or embodied AI data pipelines is a plus.
Health insurance
Vision care
Dental coverage
401(k)
Paid holidays
PTO (Paid Time Off)
Sick leave
Skills Required
- MS or PhD in Computer Science, Electrical Engineering, or a related technical field, or equivalent practical experience
- 3+ years of hands-on experience in computer vision or multimodal machine learning, including direct experience with video understanding
- Strong proficiency in Python and PyTorch
- Solid software engineering fundamentals
- Hands-on experience with CLIP or similar vision-language/video embedding models for retrieval or representation learning
- Experience building or fine-tuning LLM-based systems for video or image understanding, such as captioning, video question answering, or summarization
- Familiarity with agentic system design, tool use, multi-step reasoning, and orchestration frameworks
- Experience with large-scale video data pipelines and vector search or retrieval infrastructure
- PhD with research focus in video understanding, multimodal learning, or vision-language models
- Experience with temporal action segmentation, action localization, or instruction-level video chunking algorithms
- Experience with egocentric video datasets or head-mounted-device captured data
- Track record of deploying production-scale video search or retrieval systems
- Experience integrating foundation or vision-language models into perception or decision-making pipelines
- Publications in top-tier computer vision or machine learning venues
- Experience with humanoid robotics or embodied AI data pipelines
What We Do
Maxinsights provides scalable, high-quality, human-centric training data for Physical AI and robotics foundation-model companies. Its end-to-end data solution combines a globally distributed capture network, customizable workflows, and quality-assurance processes to make training data faster and more reliable to obtain. The company’s mission is to build the global data foundation for Physical AI by delivering data at scale with unmatched quality.
.png){kind=logo}
