On-Device Multimodal Reasoning Architect

Apple Inc.; Sunnyvale (CA); Hybrid; USD 150,000 - 278,000; Full time

Job summary

Apple Inc. in Sunnyvale, CA seeks an Applied Researcher specializing in multimodal reasoning for on-device AI systems.

You will design compact vision-language models, compress reasoning traces, and optimize decoding for real-time performance within strict device budgets. You will collaborate with hardware, compiler, and ML research teams to push on-device intelligence across Apple products, balancing latency, memory, and power while supporting privacy-preserving experiences for hundreds of

Qualifications

• MS in Computer Science, ML, AI, CV, or related field.
• Strong DL background with LLM or VLM training, post-training, or inference optimization.
• Experience with multimodal models in resource-constrained environments.
• Proficiency in Python and modern DL frameworks; familiarity with inference toolchains.

Responsibilities

• Design, train, and post-train compact vision-language models (under ~10B) for multi-step visual reasoning on-device.
• Research efficient reasoning techniques: compressed and latent chain-of-thought and budget-aware reasoning.
• Own the decoding stack for on-device inference: speculative decoding and KV-cache compression.
• Build reasoning over real-time perception experts: enable VLM to use on-device 3D reconstruction, pose, and detectors.
• Develop structured-output fusion and expert scheduling for geometry and detections within a limited context.
• Apply reasoning-focused post-training: distillation, RL methods, verifier-guided decoding.
• Drive visual token efficiency and feature compression: token pruning, adaptive resolution, connectors.
• Develop distillation, pruning, and quantization to preserve fidelity at reduced sizes.

Skills

Multimodal reasoning

On-device ML

Python

PyTorch

Education

MS in Computer Science / ML / AI / CV or related field

Tools

vLLM

llama.cpp

MLX

CoreML

Job description

Apple Inc. in Sunnyvale, CA seeks an Applied Researcher specializing in multimodal reasoning for on-device AI systems.

You will design compact vision-language models, compress reasoning traces, and optimize decoding for real-time performance within strict device budgets. You will collaborate with hardware, compiler, and ML research teams to push on-device intelligence across Apple products, balancing latency, memory, and power while supporting privacy-preserving experiences for hundreds of