Pear VC
About NomadicML
Americans drive over 5 trillion miles a year, more than 500 billion of them recorded. Buried in that footage is the next frontier of machine intelligence. At NomadicML, we’re building the platform that unlocks it. Our Vision-Language Models (VLMs) act as the new “hydraulic mining” for video, transforming raw footage into structured intelligence that powers real-world autonomy and robotics. We partner with industry leaders across self-driving, robotics, and industrial automation to mine insights from petabytes of data that were once unusable. NomadicML was founded by
Mustafa Bal
and
Varun Krishnan , who met at
Harvard University
while studying Computer Science. Mustafa
is a core contributor to
ONNX Runtime
and
DeepSpeed
with deep expertise in distributed systems and large-scale model training infrastructure
Varun
is an INFORMS Wagner Prize Finalist for his research in large-scale driver navigation AI models and one of the top chess players in the US.
Our team has built mission-critical AI systems at
Snowflake, Lyft, Microsoft, Amazon, and IBM Research , holds top-tier publications in VLMS and AI at conferences like
CVPR , and moves with the speed and clarity of a startup obsessed with impact. About the Role
We’re seeking a
Machine Learning Engineer
who thrives at the frontier of
foundation-model research and production engineering . You’ll help define how machines learn from motion: training and fine-tuning large-scale
Vision-Language Models
to reason about complex, real-world video. Your work will involve building multi-modal architectures that perceive, localize, and describe motion events (turns, lane changes, interactions, anomalies) across millions of frames, and turning those breakthroughs into robust APIs and SDKs used by enterprise customers. You’ll work directly with the founders to: Train and evaluate
VLMs specialized for motion understanding
in autonomous-driving and robotics datasets.
Design and scale
GPU-accelerated pipelines
for training, fine-tuning, and inference on multi-modal data (video + language + sensor metadata).
Build
agentic evaluation frameworks
that benchmark spatiotemporal reasoning, localization accuracy, and narrative consistency.
Develop and productionize
curation loops
that use our own models to generate and refine datasets (“AI training AI”).
Publish high-impact research (e.g., NeurIPS, CVPR) while shipping features that customers use immediately.
You’ll Excel If You Have
Strong proficiency in
Python ,
PyTorch , and large-scale ML workflows.
Research experience in
foundation models, VLMs, or multi-modal learning
(publications/patents a plus).
Ability to iterate
quickly and autonomously , running experiments end-to-end.
Experience training or fine-tuning models on
video or sensor data .
Understanding of
retrieval systems, embeddings, and GPU optimization .
Nice to Have
Contributions to open-source ML frameworks (e.g., DeepSpeed, Hugging Face).
Experience with
vector databases ,
distributed training , or
ML orchestration systems
(e.g., Ray, Kubeflow, MLflow).
Prior exposure to
autonomous-driving or robotics
datasets.
#J-18808-Ljbffr
Americans drive over 5 trillion miles a year, more than 500 billion of them recorded. Buried in that footage is the next frontier of machine intelligence. At NomadicML, we’re building the platform that unlocks it. Our Vision-Language Models (VLMs) act as the new “hydraulic mining” for video, transforming raw footage into structured intelligence that powers real-world autonomy and robotics. We partner with industry leaders across self-driving, robotics, and industrial automation to mine insights from petabytes of data that were once unusable. NomadicML was founded by
Mustafa Bal
and
Varun Krishnan , who met at
Harvard University
while studying Computer Science. Mustafa
is a core contributor to
ONNX Runtime
and
DeepSpeed
with deep expertise in distributed systems and large-scale model training infrastructure
Varun
is an INFORMS Wagner Prize Finalist for his research in large-scale driver navigation AI models and one of the top chess players in the US.
Our team has built mission-critical AI systems at
Snowflake, Lyft, Microsoft, Amazon, and IBM Research , holds top-tier publications in VLMS and AI at conferences like
CVPR , and moves with the speed and clarity of a startup obsessed with impact. About the Role
We’re seeking a
Machine Learning Engineer
who thrives at the frontier of
foundation-model research and production engineering . You’ll help define how machines learn from motion: training and fine-tuning large-scale
Vision-Language Models
to reason about complex, real-world video. Your work will involve building multi-modal architectures that perceive, localize, and describe motion events (turns, lane changes, interactions, anomalies) across millions of frames, and turning those breakthroughs into robust APIs and SDKs used by enterprise customers. You’ll work directly with the founders to: Train and evaluate
VLMs specialized for motion understanding
in autonomous-driving and robotics datasets.
Design and scale
GPU-accelerated pipelines
for training, fine-tuning, and inference on multi-modal data (video + language + sensor metadata).
Build
agentic evaluation frameworks
that benchmark spatiotemporal reasoning, localization accuracy, and narrative consistency.
Develop and productionize
curation loops
that use our own models to generate and refine datasets (“AI training AI”).
Publish high-impact research (e.g., NeurIPS, CVPR) while shipping features that customers use immediately.
You’ll Excel If You Have
Strong proficiency in
Python ,
PyTorch , and large-scale ML workflows.
Research experience in
foundation models, VLMs, or multi-modal learning
(publications/patents a plus).
Ability to iterate
quickly and autonomously , running experiments end-to-end.
Experience training or fine-tuning models on
video or sensor data .
Understanding of
retrieval systems, embeddings, and GPU optimization .
Nice to Have
Contributions to open-source ML frameworks (e.g., DeepSpeed, Hugging Face).
Experience with
vector databases ,
distributed training , or
ML orchestration systems
(e.g., Ray, Kubeflow, MLflow).
Prior exposure to
autonomous-driving or robotics
datasets.
#J-18808-Ljbffr