AI Software Engineer Intern

About This Role

• Intel's Data Center Network Edge AI team builds the AI software stack that powers next-generation workloads across Intel® Xeon® processors, Intel® dGPU based on Intel Edge AI platforms.
• We deliver the algorithms, frameworks, and optimizations that bring large-scale multi-modal intelligence to Intel customers worldwide.
• We are looking for an intern to contribute to our research and productization of Vision-Language Models (VLM) and Vision-Language-Action (VLA) models.
• You will work alongside senior software engineers to advance state-of-the-art multi-modal and embodied AI, while ensuring these models run efficiently on Intel hardware.
• Responsibilities Conduct applied research on VLM / VLA architectures, pre-training, fine-tuning, and alignment techniques (SFT, RLHF, DPO, GRPO); reproduce and extend recent work such as OpenVLA, RT-2, π05, PaLI-Gemma, Qwen-VL, and InternVL.
• Design and implement multi-modal data pipelines for cleaning, synthesis, and augmentation of image-text-action datasets.
• Investigate efficient fusion strategies between vision encoders (ViT, SigLIP, DINOv2) and language backbones (LLaMA, Qwen, Mistral), including connector design and visual token compression.
• Explore VLA-specific components such as action heads (discrete tokenization, diffusion policy, flow matching), long-horizon planning, and closed-loop control.
• Apply model optimization techniques — quantization (INT8 / FP8 / INT4, AWQ, GPTQ, SmoothQuant), pruning, distillation, KV-cache optimization, and speculative decoding — to enable efficient deployment on Intel platforms.
• Collaborate with infrastructure engineers to deploy and benchmark optimized models across Intel® Xeon®, Intel® Gaudi®, Intel® Arc™ GPU, and Intel® Core™ Ultra (NPU) targets.
• Evaluate models against industry benchmarks (MMBench, MME, SEED-Bench, LIBERO, SimplerEnv) and contribute to internal evaluation suites.
• Publish results in technical reports and, where appropriate, top-tier venues (NeurIPS, ICML, ICLR, CVPR, CoRL, RSS).