AI Researcher, Core ML (Turbo)

About the Role The Turbo team sits at the intersection of efficient inference (algorithms, architectures, engines) and post‑training / RL systems. We build and operate the systems behind Together’s API, including high‑performance inference and RL/post‑training engines that can run at production scale. Our mandate is to push the frontier of efficient inference and RL‑driven training: making models dramatically faster and cheaper to run, while improving their capabilities through RL‑based post‑training (e.g., GRPO‑style objectives). This work lives at the interface of algorithms and systems: asynchronous RL, rollout collection, scheduling, and batching all interact with engine design, creating many knobs to tune across the RL algorithm, training loop, and inference stack. Much of the job is modifying production inference systems—for example, SGLang‑ or vLLM‑style serving stacks and speculative decoding systems such as ATLAS—grounded in a strong understanding of post‑training and inference theory, rather than purely theoretical algorithm design. You’ll work across the stack—from RL algorithms and training engines to kernels and serving systems—to build and improve frontier models via RL pipelines. People on this team are often spiky: some are more RL‑first, some are more systems‑first. Depth in one of these areas plus appetite to collaborate across (and grow toward more full‑stack ownership over time) is ideal. Requirements We don’t expect anyone to check every box below. People on this team typically have deep expertise in one or more areas and enough breadth (or interest) to work effectively across the stack. The closer you are to full‑stack (inference + post‑training/RL + systems), the stronger the fit—but being spiky in one area and eager to grow is absolutely okay. You might be a good fit if you: Have strong expertise in at least one of the following, and are excited to collaborate across (and grow into) the others: Systems‑first profile: Large‑scale inference systems (e.g., SGLang, vLLM, FasterTransformer, TensorRT, custom engines, or similar), GPU performance, distributed serving. RL‑first profile: RL / post‑training for LLMs or large models (e.g., GRPO, RLHF/RLAIF, DPO‑like methods, reward modeling), and using these to train or fine‑tune real models. Model architecture design for Transformers or other large neural nets. Distributed systems / high‑performance computing for ML. Are comfortable working from algorithms to engines: Strong coding ability in Python Experience profiling and optimizing performance across GPU, networking, and memory layers. Able to take a new sampling method, scheduler, or RL update and turn it into a production‑grade implementation in the engine and/or training stack. Have a solid research foundation in your area(s) of depth: Track record of impactful work in ML systems, RL, or large‑scale model training (papers, open‑source projects, or production systems). Can read new RL / post‑training papers, understand their implications on the stack, and design minimal, correct changes in the right layer (training engine vs. inference engine vs. data / API). Operate well as a full‑stack problem solver: You naturally ask: “Where in the stack is this really bottlenecked?” You enjoy collaborating with infra, research, and product teams, and you care about both scientific quality and user‑visible wins. Minimum qualifications 3+ years of experience working on ML systems, large‑scale model training, inference, or adjacent areas (or equivalent experience via research / open source). Advanced degree in Computer Science, EE, or a related field, or equivalent practical experience. Demonstrated experience owning complex technical projects end‑to‑end. If you’re excited about the role and strong in some of these areas, we encourage you to apply even if you don’t meet every single requirement. Responsibilities Advance inference efficiency end‑to‑end Design and prototype algorithms, architectures, and scheduling strategies for low‑latency, high‑throughput inference. Implement and maintain changes in high‑performance inference engines (e.g., SGLang‑ or vLLM‑style systems and Together’s inference stack), including kernel backends, speculative decoding (e.g., ATLAS), quantization, etc. Profile and optimize performance across GPU, networking, and memory layers to improve latency, throughput, and cost. Unify inference with RL / post‑training Design and operate RL and post‑training pipelines (e.g., RLHF, RLAIF, GRPO, DPO‑style methods, reward modeling) where 90+% of the cost is inference, jointly optimizing algorithms and systems. Make RL and post‑training workloads more efficient with inference‑aware training loops—for example, async RL rollouts, speculative decoding, and other techniques that make large‑scale rollout coll