About Prime Labs
We are a small research lab building foundation models for AI-native education products, starting with English language learning. Our mission is to empower hundreds of millions of children worldwide with AI that understands how they actually speak, identifies and corrects pronunciation errors, provides phoneme-level feedback, adapts to their proficiency, and knows how to help them make progress. Our models are trained on millions of hours of proprietary, one-of-a-kind data from authentic one-on-one lessons between teachers and children, collected over many years of teaching at massive scale and analyzed with the rigor of a research lab. We are backed by DST Global, Reach Capital, K3 Ventures, and others.
About the Team
Our research team brings together AI researchers, learning scientists, veteran educators, and Stanford faculty with a remarkable breadth of experience, from scaling a business to $1B in revenue, to developing the open-source VibeVoice model at Microsoft, developing TeachLM, and working on large-scale scientific computations at Harvard and MIT.
About the Role
As a Research Engineer at Prime Labs, you'll build the systems that turn millions of hours of real lessons into audio models that teach. You'll train and iterate on models across our audio stack, from speech recognition and pronunciation assessment to teacher-like speech synthesis and real-time conversation, and build the data pipelines, training infrastructure and evaluation tooling that let us go from research idea to a model in front of children quickly.
This role is for someone who takes full ownership, moves with urgency and is comfortable when the problem isn't fully defined yet. You'll work side by side with our research scientists and engineers in our Palo Alto office, in a small, high-trust team where everyone does both research and engineering.
What You'll Do
Build and scale our audio data pipelines: quality filtering, transcription alignment and privacy-preserving processing across millions of hours of lessons
Implement, train and iterate on audio models, from ASR and pronunciation assessment to speech synthesis and speech-to-speech
Build the training infrastructure, evaluation frameworks and tooling that let us run experiments quickly and reproducibly
Optimize models for low-latency, real-time inference so conversations with children feel natural
Turn research results into production systems that ship to real learners, and take ownership of the problems you work on
What You'll Bring
Hands-on experience in ML engineering or research engineering, in industry or PhD-level research
Strong Python and PyTorch (or JAX), and the ability to implement models cleanly from papers
Experience running ML experiments end to end: data processing, training, evaluation and iteration
Experience working with large datasets and building robust, scalable data pipelines
Comfort in a fast-moving, ambiguous environment where defining the problem is part of the job
The ability to explain clearly what you tried, what worked, and why
Nice to Have
Experience with speech or audio models: ASR, TTS, speech-to-speech, neural audio codecs or speech-language models
Familiarity with autoregressive, diffusion or flow-matching generative models
Experience with distributed training, mixed precision, quantization or inference optimization
Experience processing large audio datasets: diarization, voice activity detection or forced alignment
A competitive programming or olympiad background
What We Offer
A one-of-a-kind educational data corpus with millions of hours of authentic teacher–child conversations
Real ownership within a small team where your work shapes the whole model
Dedicated GPU capacity for large-scale training
Impact you can see: your model teaches real children within months
Competitive base salary alongside an attractive equity package
Logistics
Location: on-site in Palo Alto, CA
Global team: we are a global team with offices in California, Budapest (Hungary) and Singapore. We collaborate in person through frequent travel to and from California.