Research Engineer

Hume AI; New York (NY); On-site; USD 120,000 - 180,000; Full time

Job summary

Hume AI is seeking a Research Engineer to develop speech-language and audio models, train at scale, and rigorously evaluate methods. You will work with the research team to turn questions into experiments, train and refine models, and build software that supports distributed training, inference, and benchmarking.

You will collaborate with customer teams to align research objectives with real-world data and ensure reproducible, high-quality results across in-house and client projects.

Qualifications

• At least two years of experience contributing to model training or fine-tuning with large-scale datasets (text, audio, image, or video).
• Strong Python and PyTorch skills, including developing, debugging, and maintaining training and evaluation code.
• Experience writing robust research systems that go beyond notebooks or prototypes.
• Strong experimental judgment: baselines, evaluations, and careful interpretation of results.
• Evidence of research ability via publications, model releases, or substantive open-source work.
• Comfort iterating quickly on uncertain directions and owning open-ended technical problems.

Responsibilities

• Investigate research questions in speech-language, audio, and multimodal ML; design experiments to test hypotheses.
• Train, fine-tune, validate, and further develop models in-house and for customer projects.
• Build reliable software for distributed model training, inference, and benchmarking; ensure reproducibility.
• Develop data workflows for large-scale datasets used in training and evaluation.
• Design evaluation protocols and benchmarks to measure quality, generalization, and robustness.
• Analyze datasets, model outputs, and failures to guide next steps in development.
• Collaborate with customer teams to define questions, understand data, and conduct training/evaluation.
• Document experimental designs, results, and technical decisions for reproducibility.

Skills

Python

PyTorch

Model training

Experimentation

Multimodal ML

Distributed systems

Research publications

Tools

Git

Jupyter

Linux

Job description

Hume AI is looking for a Research Engineer to advance speech-language and audio models through hands-on model development, large-scale training, and rigorous evaluation. Contribute to our endeavor to ensure that AI is guided by human values, the most pivotal challenge (and opportunity) of the 21st century.

About Us

Hume AI is a Series B startup dedicated to building artificial intelligence that is directly optimized for human well-being. As the first company to release speech language models, we’re focused on expanding our research to encompass audio understanding models and evaluation platforms for enterprises.

Our goal is to enable a future in which technology draws on an understanding of human emotional expression to better serve human goals. As part of our mission, we also conduct groundbreaking scientific research, publish in leading scientific journals like Nature, and support a non-profit, The Hume Initiative, that has released the first concrete ethical guidelines for empathic AI (www.thehumeinitiative.org). You can learn more about us on our website (https://hume.ai/) and read about us in WIRED, Forbes, and Venturebeat.

About the Role

As a Research Engineer, you will work with our research team to develop audio models and the tools used to study and improve them. You will turn open questions into experiments, train and refine models, and build the software that supports distributed training, inference, and benchmarking. Your work will connect model development with the data and evaluation systems needed to test ideas at scale.

Your work will span models developed in-house at Hume and models developed by our customers. You will collaborate directly with customer research and technical teams to understand their objectives, design research programs, and carry out model training, validation, evaluation, and continued development. Findings from these projects will inform new experiments and improvements to the models and methods we build.

This is a hands-on individual contributor role for someone who enjoys both open-ended investigation and building the code that makes research reproducible and useful. You will own technical work from an initial research question through experiments, analysis, and validated model improvements.

What You’ll Do

• Investigate research questions in speech-language, audio, and multimodal machine learning, and design experiments to test hypotheses about model behavior and capabilities.
• Train, fine-tune, validate, and further develop models built in-house at Hume and models developed by our customers.
• Build reliable software for distributed model training, inference, and benchmarking, making experiments reproducible and results comparable.
• Develop data workflows for collecting, storing, preprocessing, and analyzing large-scale datasets used in model training and evaluation.
• Design evaluation protocols and benchmarks that measure model quality, generalization, robustness, and progress against research objectives.
• Analyze datasets, model outputs, and failure cases; compare training approaches and use the evidence to guide the next round of model development.
• Collaborate with customer research and technical teams to define research questions, understand their models and data, and carry out training, validation, evaluation, and further development.
• Document experimental designs, results, limitations, and technical decisions so Hume colleagues and customer research teams can reproduce and build on the work.

What You’ll Bring

• At least two years of experience contributing to model training or fine-tuning with large-scale text, audio, image, or video datasets, with relevant depth in speech, audio, or multimodal machine learning.
• Strong Python and PyTorch skills, including the ability to develop, debug, and maintain model training and evaluation code.
• Experience writing robust, maintainable code for research systems that operate beyond an initial notebook or prototype.
• Strong experimental judgment: formulating hypotheses, establishing useful baselines, designing evaluations, and interpreting results carefully.
• Evidence of research and model-building ability through relevant publications, model releases, published systems, or substantive open-source contributions.
• Comfort iterating quickly on uncertain research directions and taking ownership of open-ended technical problems.
• Clear written and verbal communication, including the ability to explain methods, findings, and limitations to research colleagues and customer technical teams.

A PhD is not required. We value demonstrated research ability and the quality of the models, experiments, and systems you have helped build.

Bonus Points

• Experience training or fine-tuning transformer models.
• Depth in text-to-speech, automatic speech recognition, speech-to-speech, prosody, emotion recognition, or audio understanding.
• Experience with post-training, preference learning, or human evaluation of generative models.
• Experience improving the efficiency of distributed training, inference, or large-scale data processing.
• Experience collaborating with external research teams on model development or evaluation.