ML Researcher

About Us

Base Compute is an AI inference lab. Our mission is to bring AGI on device. We believe in a world where everyone has access to intelligence: fast, private and always available on your device.

We’re building the infrastructure for the next generation of on-device AI, from silicon-level optimizations to distributed inference systems.

We’re working on hard problems at the intersection of inference efficiency, model intelligence and autonomous research.

The Role

We’re looking for a Founding ML Researcher to work at the frontier of on-device AI. This role is for someone who identifies problems and potentials, designs and executes experiments and derives insights that translate into real-world performance.

You’ll have significant ownership over our research agenda and direct influence on the technical bets the company makes.

What You’ll Work On

• Inference research: Identifying and validating new approaches to on-device efficiency, including speculative decoding variants, novel quantization schemes and entirely new techniques yet to be discovered

• Model routing research: Building the intelligence that decides how requests are served between on-device vs. frontier API models

• Autoresearch pipelines: Designing systems that can autonomously explore, hypothesize and evaluate research ideas that accelerate our R&D loop

• Evaluations and benchmarks: Developing rigorous evals that measure performance in the real world, outside of clean academic settings

What We’re Looking For

• PhD in ML or equivalent industry research experience

• Deep understanding of LLM architectures and the principles of AI inference

• Expertise in a relevant topic, such as speculative decoding, quantization theory, model distillation, reinforcement learning

• A track record of producing results that people build on: research papers, open-source projects or blog posts that prove out novel ideas

• Good communication: the ability to explain complex ideas simply, give honest feedback and document findings in a reproducible way

• Nice-to-haves:

  • Familiarity with GPU and accelerator architectures and kernel optimization (CUDA, ROCm, Metal, Triton, etc.)

  • Experience deploying models under on-device constraints (memory bandwidth, latency budgets, and thermal and power ceilings)