Staff Machine Learning Engineer (Computer Vision)

Staff Machine Learning Engineer, Computer Vision

About Pravah

Pravah is building foundational intelligence for the electric grid. We apply modern machine learning to complex physical infrastructure problems spanning grid operations, weather, and geospatial systems.

Our work sits at the intersection of computer vision, physical systems, and large-scale ML, with deployments across utilities in the United States and India. We leverage multimodal data - including satellite imagery, LiDAR, and street-level data - to build high-fidelity representations of grid assets and their surroundings.

We are backed by Khosla Ventures, Pear VC, and Conviction.

To know more about who we are, what we are building, and why we are excited read this Notion!
https://pravah.notion.site/

The Role

We are hiring a Staff Machine Learning Engineer (Computer Vision) to lead the development of core perception and mapping systems for electric grid infrastructure.

This is a high-ownership, high-ambiguity role focused on building systems that operate on large-scale, heterogeneous visual data. You will define technical direction, make key architectural decisions, and ship models that are deployed in production.

Beyond core CV work, you will explore how state-of-the-art generative and vision architectures (e.g., ViTs, diffusion, flow matching) can be adapted to adjacent domains such as weather and spatiotemporal modeling.

There is also an opportunity to contribute to frontier work suitable for publication, particularly in areas where existing methods do not translate cleanly to real-world physical systems.

What You’ll Work On

Grid Mapping & Infrastructure Understanding

• Build and deploy models for object detection, segmentation, and instance-level understanding of grid assets and surroundings

• Work across satellite, aerial, street view, and LiDAR data to create unified representations of physical infrastructure

• Develop systems for depth estimation and spatial reasoning in complex real-world environments
  

Multimodal & Foundation Models

• Adapt and fine-tune vision transformers and related architectures for domain-specific tasks

• Build multimodal systems that combine visual, spatial, and structured data
  Design representations that generalize across geographies, data sources, and operating conditions
  

Applied Research & New Directions

• Bring state-of-the-art CV methods into production, bridging research and real-world deployment

• Explore the use of modern generative and vision architectures in weather and geospatial modeling applications

• Identify, prototype, and validate new approaches for modeling physical systems from visual data
  

Who You Are

• 6+ years of experience building and deploying machine learning systems, with a focus on computer vision

• Strong expertise in modern deep learning approaches, including:

  • Object detection and segmentation

  • Vision transformers and/or generative modeling approaches

  • Multimodal learning

• Proven track record of taking systems from research to production at scale

• Strong engineering fundamentals and proficiency in Python and ML frameworks (e.g., PyTorch)

• Experience working with large-scale, real-world datasets (e.g., geospatial, satellite, LiDAR) is a strong plus

• Comfortable operating in ambiguous environments, reasoning from first principles, and driving technical direction with high ownership

• Demonstrated ability to produce high-quality technical work, whether through systems, research, or publications

What You’ll Gain

• Ownership of core perception and mapping systems deployed in real-world grid operations

• Opportunity to work on hard, open-ended problems at the intersection of AI and physical infrastructure

• Ability to shape technical direction and contribute to frontier ML work

• Close collaboration with a deeply technical founding team

Why This Role

This role sits at the frontier of applying modern computer vision to large-scale physical systems. Many of the problems we work on do not have established benchmarks or standard solutions. You will operate in settings where data is heterogeneous, ground truth is incomplete, and progress requires both technical depth and first-principles thinking.