Build for what people will invent.

What you will do

• Design pre-training or post-training experiments that improve capability, context use, and computational efficiency.
• Build data preparation and training pipelines, and investigate how data quality and sampling affect model behavior.
• Develop evaluations for instruction following, reasoning, tool use, and multi-turn tasks, including cases where aggregate scores conceal failures.
• Use controlled experiments and ablations to distinguish reproducible improvements from noise or changes in evaluation conditions.
• Work with inference engineers to understand how compression and quantization affect the capabilities you train.
• Document findings and turn successful experiments into checkpoints, evaluation tools, and methods others can build on.

What we are looking for

• Strong Python skills and practical experience with PyTorch or a comparable deep-learning framework.
• Experience training, adapting, or deeply investigating language models, with work you can discuss in detail.
• A solid understanding of optimization, model architectures, and experimental design.
• The ability to debug research code, investigate unexpected results, and explain the evidence behind a decision.
• Careful engineering habits and the willingness to work across model research and deployment.

Additional experience we value

• Experience with distributed training, data quality, distillation, or model compression.
• Research on long-context modeling, preference learning, or adaptation to particular tasks and users.

What you will do

• Profile model loading, prefill, decoding, and streaming output across Apple and Android devices and desktop hardware.
• Optimize kernels, execution graphs, memory allocation, and cache management for target processors.
• Develop mixed-precision and quantized execution paths, checking numerical correctness and model quality alongside performance.
• Measure sustained behavior under realistic context lengths, concurrent workloads, and thermal conditions.
• Build repeatable benchmarks and regression tests for latency, memory, energy use, and reliability.
• Collaborate on model and runtime design, and package improvements into maintainable interfaces other engineers can use.

What we are looking for

• Strong systems programming skills in C++, Rust, or a comparable language, with Python for experimentation and tooling.
• An understanding of transformer inference, memory hierarchies, and the factors that limit compute or bandwidth utilization.
• Experience with inference runtimes, hardware acceleration, or deploying machine-learning models on constrained hardware.
• The ability to use profilers and measurements to explain a bottleneck and verify an improvement.
• Careful debugging and testing habits, including attention to numerical changes and performance regressions.

Additional experience we value

• Experience with Metal, MLX, Core ML, CUDA, or mobile GPU and NPU execution.
• Work on low-bit quantization, custom kernels, compiler optimizations, or streaming audio systems.

What you will do

• Design and train models for speech generation, audio representation, or conversational behavior, drawing on your area of expertise.
• Build audio data pipelines for cleaning, segmentation, alignment, and quality assessment.
• Investigate how dialogue history shapes prosody, delivery, and consistency across turns.
• Develop turn-taking, backchannel, and interruption behavior, including recovery when a person continues or changes their request.
• Combine listening studies with objective evaluations of intelligibility, voice quality, timing, and interaction failures.
• Improve streaming generation with systems engineers, measuring responsiveness and quality together on target devices.

What we are looking for

• Strong Python skills and hands-on experience with a modern deep-learning framework.
• Depth in at least one area of speech or audio modeling, such as synthesis, recognition, audio representations, or speech-to-speech systems.
• Experience preparing audio data and designing experiments that reveal why a model succeeds or fails.
• The ability to connect model training with inference, evaluation, and an interactive prototype.
• Attentive listening and clear technical judgment about differences conventional benchmarks may miss.

Additional experience we value

• Experience with streaming architectures, neural audio codecs, waveform generation, or full-duplex systems.
• Work on multilingual speech, expressive synthesis, human preference studies, or low-latency deployment.

What you will do

• Design APIs and SDKs for model calls, streaming, tool use, and session control, with clear behavior and useful errors.
• Build agent execution across on-device models and frontier services, including cancellation, retries, recovery, and handoffs.
• Define how applications connect tools and data sources, keeping access permissions explicit and enforceable.
• Develop tracing, evaluation hooks, and tests that help developers investigate failures as models and integrations change.
• Maintain releases, compatibility, documentation, and working examples so developers can adopt new capabilities with confidence.
• Work with model and systems engineers to improve latency and resource use throughout the application.

What we are looking for

• Experience shipping and maintaining developer tools, APIs, SDKs, or infrastructure used by other engineers.
• Strong programming skills in Python or TypeScript, with the ability to work across unfamiliar components.
• A sound understanding of asynchronous execution, API contracts, testing, and common network and process failures.
• The ability to turn an unclear developer need into a usable interface and explain the tradeoffs behind it.
• Careful judgment around permissions, sensitive data, and the consequences of tool execution.

Additional experience we value

• Experience with agent frameworks, persistent sessions, or on-device runtime integration.
• Work on open-source libraries, developer documentation, or versioned public APIs.

What you will do

• Study how people and organizations approach real tasks, and identify where AI can support their judgment and invention.
• Develop product concepts, task flows, and interactive prototypes, then test them with the people who would use them.
• Design how people follow ongoing work, interrupt speech, correct an interpretation, and approve or stop an action.
• Make system limits and choices understandable when they affect privacy, control, or a person’s ability to complete a task.
• Develop shared interaction patterns and components, carrying accessibility and visual consistency across devices.
• Work with engineers through implementation, inspect the finished interactions, and refine the experience using observed behavior.

What we are looking for

• A portfolio of product work that explains the problem, your contribution, and how the design developed through implementation.
• Strong interaction design, information architecture, and visual craft.
• Experience learning directly from users and turning observations into product decisions.
• The ability to prototype behaviors that unfold over time, including waiting, interruption, errors, and recovery.
• Clear communication and the curiosity to understand how technical choices affect the experience you are designing.

Additional experience we value

• Experience designing voice, multimodal, or other interactions that extend beyond a conventional screen.
• Interactive prototyping in code, or work on AI products, developer tools, and complex workflows.