New opportunity

Senior Deep Learning Research Engineer – Multimedia GPU team

NVIDIA · Bengaluru, India

About this role

NVIDIA is looking for a visionary Senior Deep Learning Research Engineer to join the Multimedia GPU team in Bengaluru, India. In this role, you will bridge the gap between deep learning research and production, pioneering the next generation of audio foundation models and solving complex multi-modal AI problems.

Responsibilities

Pioneer Generative AI - Audio Architecture: Lead the research, design, and development of state-of-the-art deep learning models solving complex problems in audio and speech domains.
Drive Future Technology Roadmaps: Define the technical vision and R&D strategy for NVIDIA’s future Audio, Speech and multimedia DL algorithms, ensuring alignment with hardware advancements.
Scale Massive Foundation Models: Train and optimize large-scale generative models using distributed training across massive GPU clusters.
Audio-Speech-Visual Fusion: Develop advanced algorithms for speech transformation, spatial audio, audio enhancements, and real-time video/audio enhancements using Deep Learning Algos.
Cross-Functional Leadership: Collaborate closely with NVIDIA Research, hardware architecture teams, and product groups to productize breakthrough technologies.
Productize and Deploy models on Edge: Develop and productize inference models on NVIDIA GPUs and Nvidia RTX Spark platforms as SDK/Microservices after optimization.
Technical Mentorship: Mentor senior engineers and scientists, foster a culture of technical excellence, and maintain high standards via code and design reviews.

Required qualifications

PhD in Computer Science, Artificial Intelligence, Applied Mathematics, or a related quantitative field.
10+ years of industry or post-doc experience directly developing advanced deep learning models for audio, image, and video processing.
Deep Mastery of Generative AI with a proven track record of working with Diffusion models, GANs, Transformers, VAEs, Neural Radiance Fields / 3D Gaussian Splatting, GRUs etc.
Strong Programming & Software Architecture with elite Python coding skills and a solid foundation in production-grade software design, scalability, and data structures.
Framework Proficiency with expert-level hands-on experience with PyTorch and deep familiarity with multi-modal data processing pipelines.
Proven Impact demonstrated by a strong portfolio of shipped high-impact commercial AI products or a stellar publication record at top-tier AI conferences.

Preferred qualifications

Deep understanding of Generative Audio/Speech Architecture and Algorithms.
Experience with large-scale distributed training frameworks on cluster architectures.
High proficiency in C++ and low-level GPU optimization tools like CUDA, cuDNN, Triton, or TensorRT.
Experience in building World Models or physics-informed neural networks for video synthesis.

For full details about this role, click “Apply on company site” to view the employer’s original posting.

Skills for this role

Deep LearningGenerative AIDiffusion modelsGANsTransformersVAEsNeural Radiance Fields3D Gaussian SplattingGRUsPythonPyTorchMegatron-LMDeepSpeedPyTorch FSDPC++CUDAcuDNNTritonTensorRT

Checking your profile…

Preparation suggestions, not the employer’s actual interview questions. Always check the original posting for current requirements.

Understand the role

What matters most

Pioneering Generative AI Audio Architectures
Lead the research, design, and development of state-of-the-art deep learning models for audio and speech domains, establishing next-generation audio foundation models.
Scaling Massive Foundation Models on GPU Clusters
Train and optimize large-scale generative models like Diffusion, Transformers, and Autoregressive models using distributed training frameworks across massive GPU clusters.
Bridging Research to Production and Edge Deployment
Develop and productize high-performance inference models and SDK/Microservices on NVIDIA GPUs and NVIDIA RTX Spark platforms following rigorous optimization.
Cross-Functional Collaboration and Technical Mentorship
Partner closely with NVIDIA Research, hardware architecture teams, and product groups while mentoring senior engineers and driving technical excellence.

Likely challenges to think through

Balancing Deep Research Innovation with Low-Level Edge Constraints
A key difficulty is designing complex multi-modal generative audio models while ensuring they can be highly optimized and deployed efficiently as microservices on edge platforms like NVIDIA RTX Spark. Prepare by reviewing low-level optimization tools such as TensorRT, Triton, and CUDA alongside high-level PyTorch design.
Scaling Distributed Training Across Massive GPU Clusters
Training large-scale foundation models introduces complex bottleneck trade-offs in communication overhead, cluster architecture, and memory scaling. Prepare by demonstrating hands-on familiarity with distributed training frameworks like Megatron-LM, DeepSpeed, or PyTorch FSDP.

Practice your answers

Technical practice

Designing Generative Audio Architectures

What to coverDiscuss how you would design a generative model (such as a Diffusion model or Transformer) specifically for audio and speech synthesis. Address how you handle high-sampling-rate spectrograms versus raw waveforms, and how you manage temporal dependencies.

Scaling Foundation Models Across Massive GPU Clusters

What to coverExplain the mechanics of distributed training strategies like Megatron-LM, DeepSpeed, or PyTorch FSDP when scaling massive foundation models. Discuss how you minimize communication overhead across nodes and handle memory bottlenecks.

Audio-Speech-Visual Multi-Modal Fusion

What to coverDescribe an approach for fusing audio, speech, and visual streams in real-time video/audio enhancement tasks. Address synchronization challenges and cross-modal attention mechanisms.

Low-Level GPU Optimization and Inference on Edge

What to coverExplain how you would optimize a trained deep learning model for deployment on edge platforms using tools like Triton, TensorRT, or custom CUDA kernels. Discuss trade-offs between precision quantization (e.g., FP16/INT8) and model accuracy.

Handling Multi-Modal Data Pipelines in PyTorch

What to coverDiscuss how you build high-performance data processing pipelines in PyTorch for multi-modal data streams involving video decoding and audio DSP. Address strategies to prevent I/O bottlenecks during distributed training.

Neural Radiance Fields and 3D Gaussian Splatting

What to coverExplain the core principles of Neural Radiance Fields (NeRFs) or 3D Gaussian Splatting and how they apply to spatial computing and digital twins. Address rendering speed and memory constraints.

Production-Grade Software Architecture and Scalability

What to coverDescribe how you ensure production-grade software design and scalability when transitioning experimental research code into robust SDKs or microservices. Focus on software patterns, data structures, and maintainability.

Evaluating Generative Audio and Speech Models

What to coverWhat metrics and evaluation protocols do you use to assess the quality, intelligibility, and realism of generative audio and speech models? Contrast objective metrics with subjective listening tests.

Behavioural practice

Bridging Deep Learning Research and Production

What to coverDescribe a situation where you had to transition a complex research prototype into a robust production-grade system. Explain your approach to balancing exploratory research with strict performance and deployment constraints.

Cross-Functional Leadership and Technical Collaboration

What to coverTell me about a time you collaborated closely with hardware architecture or core product teams to align software algorithms with hardware advancements. How did you resolve conflicting priorities?

Technical Mentorship and Code Review Standards

What to coverShare an example of how you mentored senior engineers and scientists to elevate team technical standards. Discuss how you handled constructive feedback during code and design reviews.

Driving Future Technology Roadmaps

What to coverDescribe a project where you defined the technical vision and R&D strategy for a complex AI domain. How did you evaluate emerging technologies and build consensus among stakeholders?

Sharpen your knowledge

Topics to prepare

Generative AI and Foundation Model Architecture
Review core architectures including Diffusion models, Transformers, and VAEs, focusing on audio and speech domains. Relevance: Directly matches the requirement to pioneer generative AI audio architecture and scale massive foundation models.
Multi-Modal Audio-Speech-Visual Fusion
Study advanced algorithms for speech transformation, spatial audio, spectrogram analysis, and audio-visual feature alignment. Relevance: Aligns with the core responsibility of developing multi-modal enhancements and real-time audio/video algorithms.
Distributed Training and Cluster Scalability
Examine methodologies for scaling model training across massive GPU clusters using distributed training strategies. Relevance: Directly ties to scaling massive foundation models and working with cluster architectures.

Skills to revise

PyTorch Framework and Multi-Modal Pipelines
Practice building custom multi-modal data processing pipelines (video decoding, audio DSP, spectrogram analysis) in PyTorch. Relevance: Essential for framework proficiency and hands-on model development as specified in the job posting.
Distributed Training Frameworks
Hands-on revision with distributed training tools such as Megatron-LM, DeepSpeed, or PyTorch FSDP. Relevance: Addresses the preferred qualification for large-scale distributed training frameworks on cluster architectures.
Low-Level GPU Optimization and C++
Review C++ programming and low-level GPU optimization tools including CUDA, cuDNN, Triton, or TensorRT. Relevance: Directly supports the preference for low-level GPU optimization and deploying models on edge/microservices.

Your preparation checklist

Prepare Research Portfolio and Publication Summary
Compile a structured portfolio highlighting your shipped commercial AI products or publications in top-tier conferences (e.g., CVPR, ICCV, SIGGRAPH, NeurIPS, ICASSP, Interspeech) focusing on audio, image, and video processing.
Build a Generative Audio and Multi-Modal Walkthrough
Prepare a detailed technical overview of a generative model or multi-modal pipeline you designed (e.g., using Diffusion models, Transformers, or VAEs for audio/speech), addressing DSP and spectrogram analysis pipelines.
Review Distributed Training and GPU Cluster Scalability
Synthesize your experience with large-scale distributed training frameworks like Megatron-LM, DeepSpeed, or PyTorch FSDP, and be ready to discuss handling memory bottlenecks and communication overhead across massive GPU clusters.
Demonstrate Low-Level GPU Optimization and C++ Proficiency
Prepare concrete code examples or architectural explanations involving C++ and low-level GPU optimization tools such as CUDA, cuDNN, Triton, or TensorRT for edge deployment and microservices.
Draft Technical Leadership and Mentorship Examples
Outline specific past experiences where you mentored senior engineers, conducted code and design reviews, and collaborated cross-functionally with hardware or product teams.
Formulate Thoughtful Employer Questions
Prepare targeted questions for the interviewing team regarding NVIDIA's multimedia GPU roadmap, collaboration between NVIDIA Research and product groups like NeMo/AI4Media, and edge deployment targets on RTX Spark platforms.

Availability last checked: 4 Nov 2026. Status can change after a check; confirm on the employer’s website. Report an issue · Our checking policy

KEEP EXPLORING

Similar AI jobs

Related skills and specializations in India. Matched to this role, not your profile.

Explore more jobs

Finding similar opportunities…