Latest AI JobsBrowse Jobs
M

RESEARCHER, EFFICIENT INFERENCE

makermaker.aiSan FranciscoTechnical Staff
Apply on company site
Salary
Not listed
Job type
Full-time
Experience
Senior · 5+ yrs
Posted
25 Sept 2026
Status
Open · checked 7d ago

About this role

The Researcher, Efficient Inference position at makermaker.ai is an on-site role in San Francisco focused on optimizing machine learning model performance and serving efficiency. This senior research position suits experienced professionals with a strong background in model compression, speculative decoding, and production-level machine learning implementation.

Role Overview

makermaker.ai seeks a Researcher, Efficient Inference to work on-site in San Francisco. The position sits at the intersection of model architecture and inference performance, requiring the creation of methods that improve accuracy, latency, and cost trade-offs. The individual will design algorithms, perform evaluations, and collaborate with engineers to deploy solutions into production.

Key Responsibilities

Research and create quantization techniques, covering post-training quantization, quantization-aware training, mixed-precision regimes, and low-bit-width arithmetic.
Develop and assess speculative decoding methods, including draft models, tree attention, parallel speculation, and lookahead decoding.
Study training-time efficiency strategies that integrate with inference, such as distillation, sparse attention, mixture-of-experts, low-rank adaptation, and pruning.
Execute controlled experiments at production scale to evaluate performance on real workloads rather than toy benchmarks.
Collaborate with the inference engineering team to transition research outcomes into operational production environments.
Review efficient machine learning literature to incorporate relevant findings into the company stack.
Share findings internally and publish research when appropriate, while coordinating with training researchers to align efficiency choices with architecture decisions.

Requirements

Demonstrated background in machine learning research focusing on efficiency methods like quantization, speculative decoding, distillation, mixture-of-experts, or sparse attention.
Minimum of five years of practical research experience.
Familiarity with training and inference performance characteristics.
Proficiency in PyTorch, Jax, or equivalent tools, with comfort working at the kernel and serving-framework level.
History of transitioning efficiency research from prototypes to production systems.
Statistical expertise for rigorous experimental evaluation and strong written communication skills.
Published research at venues such as NeurIPS, ICML, ICLR, MLSys, or comparable locations.

Nice to Have

PhD in machine learning, systems, or a related academic discipline.
Contributions to open-source libraries related to quantization, speculative decoding, or efficient inference.
Familiarity with hardware-aware optimization and accelerator-specific tooling.
Experience in numerical methods, low-precision arithmetic, or approximate computation.

Skills and Tools

PyTorch
Jax
Quantization methods
Speculative decoding
Mixture-of-Experts
Sparse and structured attention
Distillation
Low-rank adaptation

About the Company

makermaker.ai is a small team based on-site in San Francisco building autonomous research agents for recursive self-improvement. These multi-agent systems propose, run, and analyze machine learning experiments.

How to Apply

Interested candidates should apply on the official company careers page.

Skills and tech stack

PyTorchJaxJAXMachine Learning

Questions candidates ask

Where is the Researcher, Efficient Inference role located?

The position is on-site with a small team based in San Francisco.

How many years of experience are required for this position?

Candidates must have 5+ years of hands-on research experience.

What programming frameworks and tools should applicants be fluent in?

Applicants should be fluent in PyTorch, Jax, or equivalent tools, and comfortable working at the kernel and serving-framework level.

Is a PhD required for this role?

A PhD in machine learning, systems, or a related field is listed as a nice-to-have qualification rather than a strict requirement.

What publishing history is expected for applicants?

Applicants should have a track record of published research at venues such as NeurIPS, ICML, ICLR, MLSys, or comparable venues.

Ready to apply?

The application is on the official makermaker.ai careers page.

Apply on company site