Software Engineer, Inference API

Cerebras Systems builds the world's largest AI chip, 56 times larger than GPUs. This architecture allows Cerebras to deliver industry-leading training and inference speeds; over 10 times faster than GPU-based hyperscale cloud inference services. This order of magnitude increase in speed is transforming the user experience of AI applications, unlocking real-time iteration and increasing intelligence via additional agentic computation. Cerebras works with the leading model labs, global enterprises, and cutting-edge AI-native startups.

OpenAI recently announced a multi-year partnership with Cerebras, to deploy 750 megawatts of scale, transforming key workloads with ultra high-speed inference.

Cerebras is building a new generation of disaggregated AI inference systems that combine GPU-accelerated prefill with ultra-fast decode on the Cerebras Wafer-Scale Engine.

We are hiring a Software Engineer to build and evolve the ML API layer that makes this heterogeneous serving system accessible, reliable, and easy to use. You will work across our inference APIs, model integration layer, request-routing services and Cerebras inference platform to deliver a consistent experience across models and accelerator backends.

This role sits at the intersection of machine learning systems, API design, model serving, and distributed systems. You will enable new model architectures and inference capabilities, define stable user-facing behavior, and ensure that features such as streaming, sampling, tool use, structured outputs, multimodal inputs, and model configuration behave correctly and consistently in production.

You will work closely with model enablement, compiler, runtime, cloud infrastructure, product, customer-facing teams, and customers directly. This is a hands‑on software engineering role for someone who enjoys turning rapidly evolving ML capabilities into durable, production-quality APIs.

• Build production ML inference APIs. Design, implement, and maintain APIs for chat completions, text generation, streaming, model configuration, tool calling, structured outputs, multimodal inputs, and other emerging inference capabilities.

• Deliver a unified serving experience. Create consistent request and response semantics across GPU prefill, Cerebras decode, and other heterogeneous inference backends.

• Enable new models and capabilities. Integrate emerging foundation models, tokenizers, prompt formats, sampling methods, attention variants, multimodal inputs, and model-specific features into the serving platform.

• Own API compatibility and evolution. Maintain compatibility with widely adopted inference interfaces while designing Cerebras‑specific extensions. Establish clear versioning, deprecation, validation, and backward compatibility practices.

• Integrate with model‑serving runtimes. Extend and integrate custom inference services with vLLM, PyTorch, Hugging Face libraries, the AMD ROCm stack, and Cerebras runtime components.

• Support disaggregated inference. Build the control and data paths required to coordinate GPU prefill with Cerebras decode, including request routing, state transfer, error handling, retries, and lifecycle management.

• Improve serving performance. Optimize streaming behavior, time to first token, request latency, throughput, batching, serialization, tokenization, scheduling, and communication between serving components.

• Ensure functional and numerical correctness. Build validation systems for tokenization, sampling, logits, generated outputs, precision changes, model upgrades, determinism, and compatibility across serving backends.

• Strengthen reliability and observability. Define end‑to‑end service indicators and build structured logging, tracing, metrics, dashboards,…