Research Engineer

Kalpa Labs; San Francisco (CA); On-site; USD 180,000 - 240,000; Full time

Job summary

Kalpa Labs is hiring a Founding ML Research Engineer to drive end-to-end development of generalist audio models. You will work across data, pre-training, post-training and evaluation, delivering research ideas to production rapidly with a small, compute-heavy team.

You’ll explore multi-modal architectures, scalable data pipelines, and high-performance ML systems. This role demands deep knowledge of large neural networks and strong software engineering skills to ship work quickly.

Qualifications

• Experience researching pre-training and post-training large neural networks.
• Strong ML systems and engineering depth (distributed training, performance, reliability).
• Ability to spec, build, debug, and ship in ambiguity.

Responsibilities

• Research and design efficient multi-modal architectures and codecs for audio.
• Post-train audio models to enable instruction following and in-context learning for text and audio.
• Build large-scale speech model pre-training and post-training pipelines (SFT/RLHF-style, distillation, etc.).
• Create scalable data compute pipelines: dataset curation, filtering, tokenization, evaluation harnesses.

Skills

ML research

Distributed training

Multimodal models

Speech/audio

Hardware acceleration

Education

PhD in ML / AI

Tools

PyTorch

TensorFlow

HuggingFace

Job description

We’re hiring a Founding ML Research Engineer to work work through the full stack across data, pre-training, post-training & evals for training Generalist Audio Models. You’ll work through the entire stack with small team, tons of compute, high autonom, and see your research ideas making it to production within a week(s).

What you’ll do

• Research better multi-modal architectures & codecs that are efficient across both spoken speech & general audio.
• Post-train audio models to have LLM like instruction following & in-context learning but over both text and audio.
• Build large-scale speech model pre-training and post-training (SFT/RLHF-style, distillation, preference optimization, etc.).
• Build scalable data compute pipelines: dataset curation, filtering, mixing, tokenization/feature pipelines, evaluation harnesses.
• Look at lots of data & hear lots of audio.

What we’re looking for

• Industry/Academia experience pre-training / post-training large neural networks; speech/audio is a plus but not required, language/vision experience is also relevant.
• Strong ML systems and engineering depth (distributed training, performance, reliability).
• Comfort operating in ambiguity: you can spec, build, debug, and ship.
• A hunger to always ask – what would the next frontier look like?