This is a fully remote position, open to applicants in Poland, +4 more countries.

📋 Description

• Establish technical connections with clients to develop multimodal AI systems aimed at document intelligence, personalization, and image/video analysis, encompassing everything from architectural design to production rollout.

• Advise clients on model training approaches across different modalities, such as layout-aware document encoders, user-item interaction models, and spatiotemporal video representations.

• Tackle visual content challenges, including issues related to image resolutions, optimizing vision encoders to meet production latency requirements, efficient video frame sampling, and temporal reasoning.

• Advocate for customer requirements to NVIDIA product teams, converting field experiences into strategic decisions for the roadmap concerning NeMo, TensorRT-LLM, Dynamo, and RAPIDS.

• Involve the developer community by organizing hackathons, delivering technical presentations, conducting demonstrations, and creating reference blueprints.

• Convert cutting-edge AI technologies into precise, production-ready solutions that provide measurable business advantages.

⛳️ Requirements

• MS or PhD in Computer Science, Engineering, or a related field will be considered.

• Over 7 years of experience in applied AI/ML.

• Practical experience in document understanding and visual content analysis.

• A proven history of constructing or enhancing VLMs and Omni models.

• Familiarity with NVIDIA's ecosystem, including TensorRT-LLM, NeMo, RAPIDS, or similar training and inference frameworks.

• Exceptional communication abilities; comfortable collaborating with research scientists, ML engineers, and business partners.

• Experience with multimodal AI systems that manage various modalities, including audio and video, and complex structures such as layouts, tables, and multi-page reasoning.

• Knowledge of retrieval and search systems, including dense retrieval, ANN indexing, and re-ranking pipelines.

• Experience in optimizing vision encoders for production through quantization, pruning, or architectural modifications.

• Published works or contributions to open-source projects in the field of multimodal learning.